Dalam beberapa tahun terakhir, dunia teknologi mengalami perkembangan serta tingkat penggunaan yang luar biasa akibat perkembangan Generative AI. Kehadiran sistem seperti ChatGPT, Gemini, Grok, Claude, Deepseek, Qwen, Mistral, dll tidak hanya mengubah cara manusia berinteraksi dengan komputer, tetapi juga menggeser batas antara alat dan “teman berpikir / diskusi”. Bahkan kini, perkembangan tidak berhenti pada generative AI saja, tetapi mulai bergerak menuju AI Agent, dimana AI mampu bertindak secara mandiri, mengambil keputusan, dan menjalankan tugas kompleks.<img src="/uploads/content/1775802327724-24162545.webp">Gambar 1. Hype Cycle for Generative AI, 2025 (Sumber : Gartner.com)Fenomena ini memunculkan satu pertanyaan besar, apakah kita sedang berada dalam fase “AI Bubble” atau bagian dari evolusi teknologi yang memang tak terhindarkan?. Hasil pemikiran ini mencoba melihat fenomena tersebut secara reflektif serta tidak hanya dari sisi teknologi, tetapi juga dari perspektif sosial dan ekonomi.Konsep “AI Bubble” tidak dapat dilepaskan dari kerangka Gartner Hype Cycle, yang menjelaskan bahwa perkembangan teknologi umumnya melalui lima tahap utama, yaitu technology trigger, peak of inflated expectations, trough of disillusionment, slope of enlightenment, dan plateau of productivity [1]. Model ini menggambarkan bagaimana ekspektasi terhadap teknologi baru cenderung meningkat secara signifikan sebelum mengalami penurunan, dan kemudian secara bertahap mencapai tingkat kematangan serta produktivitas yang stabil.<img src="/uploads/content/1775802366159-312163114.webp"> Gambar 2. Gartner Hype Cycle (Sumber : Forbes.com)Menariknya, beberapa penelitian menunjukkan bahwa generative AI berada di fase “peak of inflated expectations” yaitu kondisi di mana ekspektasi publik sangat tinggi, sering kali melampaui realitas teknologi [2]. Bahkan studi terbaru tentang tingkat pemaakian generative AI menunjukkan bahwa teknologi ini tidak hanya mengikuti siklus teknis, tetapi juga siklus emosional yang dimulai dari rasa antusiasme, kebingungan, hingga adaptasi [3]. Dalam konteks ini, AI bubble bukan sekadar fenomena ekonomi, tetapi juga fenomena sosial.Generative AI berkembang dengan kecepatan yang jarang terjadi dalam sejarah teknologi. Menurut laporan Stanford University, investasi global pada generative AI mencapai sekitar $33.9 miliar dan terus meningkat, sementara adopsi AI dalam organisasi melonjak dari 55% menjadi 78% hanya dalam satu tahun [4]. <img src="/uploads/content/1775802388991-400756143.webp">Gambar 3. Perusahaan Global dengan Investasi Terbesar dalam Pengembangan Artificial Intelligence (Sumber : Statista.com) Namun, kenaikan ini tidak hanya terjadi pada level teknologi, tetapi juga pada ekspektasi publik yang berkembang seperti : • AI akan menggantikan sebagian besar pekerjaan manusia • AI mampu berpikir seperti manusia • AI akan menjadi solusi universal untuk berbagai masalahPadahal, dalam praktiknya, AI masih memiliki keterbatasan mendasar seperti bias data, kesalahan (hallucination) dan ketergantungan pada input yang diberikan. Fenomena ini menunjukkan bahwa kondisi saat ini tidak hanya sedang menghadapi inovasi teknologi, tetapi juga gelombang persepsi kolektif yang sangat kuat. Realitas di Balik Hype: Antara Harapan dan Kekecewaan Meskipun hype sangat tinggi, pada kenyataannya implementasi AI sering kali tidak seindah ekspektasi. Beberapa laporan menunjukkan bahwa banyak proyek AI belum memberikan dampak signifikan dalam produktivitas atau profit. Hal ini memperkuat indikasi bahwa pasa masa ini mulai memasuki fase “disillusionment”, di mana organisasi mulai menyadari bahwa implementasi AI tidak semudah yang dibayangkan. Bahkan, banyak perusahaan menghadapi tantangan seperti: • Integrasi sistem yang kompleks • Kualitas data yang tidak memadai • Kebutuhan adaptasi organisasi • Risiko etika dan keamananFenomena ini sejalan dengan studi yang menyebutkan bahwa gelombang hype teknologi sering kali diikuti oleh fase koreksi sebelum mencapai kematangan [5]. Namun, penting untuk dicatat bahwa koreksi tidak berarti kegagalan melainkan proses menuju stabilitas.AI Agent: Evolusi Berikutnya atau Bubble Baru? Jika generative AI berfokus pada “menciptakan”, maka AI Agent melangkah lebih jauh yaitu “bertindak”. AI Agent adalah sistem yang tidak hanya menghasilkan output, tetapi juga dapat mengambil keputusan, menjalankan tugas berulang serta berinteraksi dengan sistem lain secara mandiri. Konsep ini membawa dampak besar terhadap masa depan pekerjaan dan organisasi. Namun, di sisi lain, AI Agent juga berpotensi menciptakan hype baru.Pertanyaannya: apakah AI Agent benar-benar siap secara teknis dan sosial? Sejarah menunjukkan bahwa setiap gelombang teknologi membawa janji besar namun tidak semuanya langsung terealisasi. Oleh karena itu, AI Agent bisa dilihat sebagai dua hal sekaligus yaitu Evolusi logis dari AI dan Potensi bubble berikutnya jika ekspektasi kembali tidak terkendali. Selanjutnya apakah ini bubble atau evolusi?. Jawaban dari pertanyaan ini tidak sesederhana “ya” atau “tidak”.Perkembangan AI saat ini memiliki fondasi yang lebih kuat dibandingkan era bubble sebelumnya seperti “dot-com”[6]. Investasi yang masuk tidak sepenuhnya dari sisi spekulatif, tetapi juga didorong oleh kebutuhan nyata dan potensi produktivitas [7]. Di sisi lain, pola hype yang terjadi sangat mirip dengan fenomena bubble dalam sejarah teknologi. Artinya, kemungkinan besar saat ini perkembangannya sedang berada dalam kondisi “hybrid”. Dimana ada elemen bubble (overhype, ekspektasi tinggi) dan elemen evolusi (teknologi nyata dan terus berkembang). Dengan kata lain, bukan teknologinya yang menjadi bubble tetapi cara manusia memaknainya.Euforia terhadap Generative AI dan AI Agent akhirnya menunjukkan bahwa yang sedang terjadi bukan sekadar kemajuan teknologi, melainkan refleksi dari harapan, ketakutan, dan ekspektasi manusia terhadap masa depan. Istilah “AI Bubble” mungkin terlalu sederhana, tetapi mengabaikan potensi overhype juga bukan tanpa risiko. Di tengah derasnya arus inovasi, yang dibutuhkan bukanlah sikap ekstrem yang berarti terlalu optimistis maupun terlalu skeptis, melainkan keseimbangan dalam melihat dan menyikapinya. AI perlu dipahami secara kritis, tanpa terjebak dalam hype, namun juga tanpa menutup diri terhadap potensinya. Pada akhirnya, arah perkembangan AI tidak hanya ditentukan oleh kecanggihan algoritma dan perangkat kerasnya, tetapi oleh pilihan manusia dalam menggunakannya yang dianggap sebagai alat, teman, atau sekadar ilusi.Referensi: [1] J. Werner, “The trough of disillusionment and four outliers on the Gartner hype cycle,” Forbes, Jul. 18, 2024. [Online]. Available: <a href="https://www.forbes.com/sites/johnwerner/2024/07/18/the-trough-of-disillusionment-and-four-outliers-on-the-gartner-hype-cycle/">https://www.forbes.com/sites/johnwerner/2024/07/18/the-trough-of-disillusionment-and-four-outliers-on-the-gartner-hype-cycle/</a> [2] R. Wandile, “Gartner places generative AI at the top of the hype cycle: What does it mean for the technology?,” LinkedIn, Sep. 12, 2023. [Online]. Available: <a href="https://www.linkedin.com/pulse/gartner-places-generative-ai-top-hype-cycle-what-does-rahul-wandile/">https://www.linkedin.com/pulse/gartner-places-generative-ai-top-hype-cycle-what-does-rahul-wandile/</a>.[3] V. Truong, “Hype and Adoption of Generative Artificial Intelligence Applications,” 2025, arXiv. <a href="https://doi.org/10.48550/arXiv.2504.18081">https://doi.org/10.48550/arXiv.2504.18081</a>.[4] Maslej, N., Fattorini, L., et al. (2025). Artificial Intelligence Index Report 2025 (Version 3). arXiv. <a href="https://doi.org/10.48550/ARXIV.2504.07139">https://doi.org/10.48550/ARXIV.2504.07139</a>.[5] Taherdoost, H. (2025). From hype to bubble: a historical analysis of technology trends and the case for artificial intelligence. Future Digital Technologies and Artificial Intelligence, 1(1), 1–6. <a href="https://doi.org/10.55670/fpll.fdtai.1.1.1">https://doi.org/10.55670/fpll.fdtai.1.1.1</a>.[6] Janus Henderson Investors, “AI versus the dotcom bubble: 8 reasons the AI wave is different,” Oct. 16, 2025. [Online]. Available: <a href="https://www.janushenderson.com/corporate/article/ai-versus-the-dotcom-bubble-8-reasons-the-ai-wave-is-different/">https://www.janushenderson.com/corporate/article/ai-versus-the-dotcom-bubble-8-reasons-the-ai-wave-is-different/</a>[7] E. Brynjolfsson, D. Rock, and C. Syverson, “Artificial Intelligence and the Modern Productivity Paradox,” The Economics of Artificial Intelligence. University of Chicago Press, pp. 23–60, 2019. <a href="https://doi.org/10.7208/chicago/9780226613475.003.0001">https://doi.org/10.7208/chicago/9780226613475.003.0001.</a>

In recent years, the tech industry has seen remarkable advancements and a surge in adoption driven by the rise of Generative AI. The emergence of systems such as ChatGPT, Gemini, Grok, Claude, Deepseek, Qwen, Mistral, and others has not only transformed how humans interact with computers but has also blurred the line between tools and “thinking/discussion partners.” Even now, advancements are not limited to generative AI alone but are shifting toward AI Agents, where AI can act autonomously, make decisions, and perform complex tasks.<img src="/uploads/content/1775802327724-24162545.webp">Figure 1. Hype Cycle for Generative AI, 2025 (Source: Gartner.com)This phenomenon raises a major question: are we currently in an “AI Bubble” phase, or part of an inevitable technological evolution? This analysis attempts to examine the phenomenon reflectively—not only from a technological perspective but also from social and economic viewpoints.The concept of the “AI Bubble” cannot be separated from the Gartner Hype Cycle framework, which explains that technological development generally goes through five main stages: technology trigger, peak of inflated expectations, trough of disillusionment, slope of enlightenment, and plateau of productivity [1]. This model illustrates how expectations regarding new technologies tend to rise significantly before declining, and then gradually reach a stable level of maturity and productivity.<img src="/uploads/content/1775802366159-312163114.webp"> Figure 2. Gartner Hype Cycle (Source: Forbes.com)Interestingly, several studies indicate that generative AI is in the “peak of inflated expectations” phase—a stage where public expectations are extremely high, often exceeding the technology’s actual capabilities [2]. Even recent studies on generative AI adoption rates suggest that this technology follows not only a technical cycle but also an emotional cycle, ranging from enthusiasm to confusion and ultimately adaptation [3]. In this context, the AI bubble is not merely an economic phenomenon but also a social one.Generative AI is evolving at a pace rarely seen in the history of technology. According to a Stanford University report, global investment in generative AI has reached approximately $33.9 billion and continues to rise, while AI adoption within organizations has surged from 55% to 78% in just one year [4]. <img src="/uploads/content/1775802388991-400756143.webp">Figure 3. Global Companies with the Largest Investments in Artificial Intelligence Development (Source: Statista.com) However, this surge is not limited to the technological level but also extends to evolving public expectations, such as :• AI will replace most human jobs• AI is capable of thinking like humans• AI will become a universal solution to various problemsIn reality, however, AI still has fundamental limitations such as data bias, errors (hallucinations), and dependence on the provided input. This phenomenon indicates that the current landscape is not only facing technological innovation but also a powerful wave of collective perception. The Reality Behind the Hype: Between Hope and Disappointment Despite the intense hype, in reality, AI implementation often falls short of expectations. Several reports indicate that many AI projects have yet to deliver significant impacts on productivity or profit. This reinforces the indication that the current phase is entering a “disillusionment” stage, where organizations are beginning to realize that AI implementation is not as straightforward as imagined. In fact, many companies face challenges such as: • Complex system integration• Inadequate data quality• Organizational adaptation needs• Ethical and security risksThis phenomenon aligns with studies noting that waves of technological hype are often followed by a correction phase before reaching maturity [5]. However, it is important to note that correction does not mean failure but rather a process toward stability.AI Agents: The Next Evolution or a New Bubble? While generative AI focuses on “creating,” AI Agents go a step further by “acting.” An AI Agent is a system that not only generates output but can also make decisions, perform repetitive tasks, and interact with other systems autonomously. This concept has significant implications for the future of work and organizations. However, on the other hand, AI Agents also have the potential to create a new wave of hype.The question is: are AI Agents truly ready, both technically and socially? History shows that every wave of technology brings great promises, yet not all are immediately realized. Therefore, AI Agents can be viewed as two things at once: the logical evolution of AI and the potential for the next bubble if expectations once again spiral out of control. So, is this a bubble or an evolution? The answer to this question is not as simple as “yes” or “no.”Current AI development has a stronger foundation compared to previous bubble eras like the “dot-com” era[6]. The investments flowing in are not entirely speculative but are also driven by real needs and productivity potential[7]. On the other hand, the current hype patterns closely resemble historical technology bubble phenomena. This suggests that the current development is likely in a “hybrid” state—combining bubble elements (overhype, high expectations) with evolutionary elements (real, continuously advancing technology). In other words, it is not the technology itself that is the bubble, but how humans interpret it.The euphoria surrounding Generative AI and AI Agents ultimately reveals that what is unfolding is not merely technological progress, but a reflection of human hopes, fears, and expectations regarding the future. The term “AI Bubble” may be overly simplistic, but ignoring the potential for overhype is not without risk. Amid the torrent of innovation, what is needed is not an extreme stance—whether overly optimistic or overly skeptical—but a balanced approach to viewing and addressing it. AI must be understood critically, without getting caught up in the hype, yet without closing oneself off to its potential. Ultimately, the direction of AI’s development is determined not only by the sophistication of its algorithms and hardware, but by human choices in using it—whether as a tool, a companion, or merely an illusion.Reference: [1] J. Werner, “The trough of disillusionment and four outliers on the Gartner hype cycle,” Forbes, Jul. 18, 2024. [Online]. Available: <a href="https://www.forbes.com/sites/johnwerner/2024/07/18/the-trough-of-disillusionment-and-four-outliers-on-the-gartner-hype-cycle/">https://www.forbes.com/sites/johnwerner/2024/07/18/the-trough-of-disillusionment-and-four-outliers-on-the-gartner-hype-cycle/</a> [2] R. Wandile, “Gartner places generative AI at the top of the hype cycle: What does it mean for the technology?,” LinkedIn, Sep. 12, 2023. [Online]. Available: <a href="https://www.linkedin.com/pulse/gartner-places-generative-ai-top-hype-cycle-what-does-rahul-wandile/">https://www.linkedin.com/pulse/gartner-places-generative-ai-top-hype-cycle-what-does-rahul-wandile/</a>.[3] V. Truong, “Hype and Adoption of Generative Artificial Intelligence Applications,” 2025, arXiv. <a href="https://doi.org/10.48550/arXiv.2504.18081">https://doi.org/10.48550/arXiv.2504.18081</a>.[4] Maslej, N., Fattorini, L., et al. (2025). Artificial Intelligence Index Report 2025 (Version 3). arXiv. <a href="https://doi.org/10.48550/ARXIV.2504.07139">https://doi.org/10.48550/ARXIV.2504.07139</a>.[5] Taherdoost, H. (2025). From hype to bubble: a historical analysis of technology trends and the case for artificial intelligence. Future Digital Technologies and Artificial Intelligence, 1(1), 1–6. <a href="https://doi.org/10.55670/fpll.fdtai.1.1.1">https://doi.org/10.55670/fpll.fdtai.1.1.1</a>.[6] Janus Henderson Investors, “AI versus the dotcom bubble: 8 reasons the AI wave is different,” Oct. 16, 2025. [Online]. Available: <a href="https://www.janushenderson.com/corporate/article/ai-versus-the-dotcom-bubble-8-reasons-the-ai-wave-is-different/">https://www.janushenderson.com/corporate/article/ai-versus-the-dotcom-bubble-8-reasons-the-ai-wave-is-different/</a>[7] E. Brynjolfsson, D. Rock, and C. Syverson, “Artificial Intelligence and the Modern Productivity Paradox,” The Economics of Artificial Intelligence. University of Chicago Press, pp. 23–60, 2019. <a href="https://doi.org/10.7208/chicago/9780226613475.003.0001">https://doi.org/10.7208/chicago/9780226613475.003.0001.</a>

AI Bubble atau Evolusi ? Euforia Generative AI dan AI Agent

AI Bubble or Evolution? The Hype Surrounding Generative AI and AI Agents

Bayangkan sebuah masa depan di mana kecerdasan buatan bukan lagi sekadar teman mengobrol, melainkan asisten yang benar-benar bisa bekerja nyata untuk Anda. Kita sering bermimpi memiliki AI yang mampu mengelola kode di GitHub hingga menyusun jadwal di kalender secara otomatis. Namun, langkah besar ini membawa kekhawatiran baru: amankah data pribadi kita? Makalah ini memperkenalkan ZENITH, sebuah sistem orkestrasi AI generasi terbaru yang dirancang untuk menjawab keraguan tersebut. Fokus utama kami bukan hanya pada kecanggihan teknologi, tetapi pada cara menjaga kedaulatan identitas pengguna di setiap tindakan yang diambil oleh sistem. Melalui ZENITH, kami mengeksplorasi tantangan dalam membangun agen AI yang kuat namun tetap aman. Kami menyadari bahwa menyimpan kunci akses secara sembarangan sangat berbahaya di dunia digital saat ini. Oleh karena itu, ZENITH menggunakan pendekatan unik yang memisahkan tugas AI dari data sensitif pengguna. Dengan mengintegrasikan sistem keamanan tingkat tinggi seperti Auth0 Token Vault dan protokol verifikasi dua langkah yang real-time, kami memastikan bahwa AI tidak pernah menyentuh sandi asli Anda. Setiap tindakan penting tetap membutuhkan persetujuan manusia sebagai kendali utama. Hasilnya, ZENITH membuktikan bahwa kita bisa memiliki AI yang sangat membantu tanpa perlu mengorbankan privasi. Ini adalah cetak biru bagi teknologi masa depan yang cerdas, aman, dan tetap memanusiakan penggunanya.

Imagine a future where artificial intelligence is no longer just a chat companion, but an assistant that can actually do real work for you. We often dream of having AI capable of managing code on GitHub and automatically organizing schedules in our calendars. However, this major leap brings new concerns: is our personal data safe? This paper introduces ZENITH, a next-generation AI orchestration system designed to address these concerns. Our primary focus is not merely on technological sophistication, but on how to safeguard user identity sovereignty in every action taken by the system. Through ZENITH, we explore the challenges of building a powerful yet secure AI agent. We recognize that carelessly storing access keys is extremely dangerous in today’s digital world. Therefore, ZENITH employs a unique approach that separates AI tasks from users’ sensitive data. By integrating high-level security systems such as Auth0 Token Vault and real-time two-factor verification protocols, we ensure that the AI never comes into contact with your actual passwords. Every critical action still requires human approval as the primary safeguard. As a result, ZENITH proves that we can have highly helpful AI without sacrificing privacy. This is a blueprint for future technology that is intelligent, secure, and remains user-centric.

Sering kali kita masih membayangkan kecerdasan buatan (AI) hanya sebatas chatbot. tempat bertanya, mencari jawaban, atau sekadar berdialog. Namun, bagaimana jika AI mampu melampaui peran tersebut dan bertindak sebagai asisten pribadi yang benar-benar mengeksekusi tugas kompleks, mulai dari menerbitkan kode ke repositori GitHub hingga mengatur jadwal dengan menggunakan Google Calendar, tanpa mengorbankan keamanan data penggunanya? Pertanyaan inilah yang menjadi fondasi dari ZENITH: sebuah orkestrator AI generasi baru yang tidak hanya berfokus pada kecerdasan, tetapi juga menempatkan kedaulatan identitas (identity sovereignty) sebagai prinsip utama dalam setiap aksi yang dijalankan.Membangun sebuah sistem AI kompleks seperti ZENITH, yang mampu mengoordinasikan model-model canggih seperti Google Gemini 3.1 Pro, Imagen, dan Veo 3.1 secara simultan, menghadirkan sebuah paradoks fundamental: bagaimana merancang sistem yang cukup kuat untuk “bertindak” (acting) atas nama pengguna, namun tetap cukup aman sehingga tidak pernah menyentuh kredensial paling sensitif milik pengguna? Dalam praktik pengembangan agen AI, kesalahan mendasar yang masih sering ditemui adalah penggunaan metode hardcoding API key atau penyimpanan kredensial di sisi frontend state yang rentan terhadap eksploitasi. Bagi ZENITH, pendekatan tersebut bukanlah sebuah opsi. Oleh karena itu, arsitektur Zero-Trust diterapkan secara menyeluruh, dengan Auth0 Token Vault sebagai fondasi utama dalam pengelolaan identitas dan akses. Pengalaman empiris menunjukkan bahwa kredensial yang terekspos dapat dieksploitasi dalam hitungan detik. Bot otomatis mampu mengonsumsi kuota API secara masif, menegaskan bahwa dalam ekosistem digital modern, akses tidak sah bukan sekadar kemungkinan, melainkan keniscayaan. Menjawab tantangan ini, pengembangan ZENITH berfokus pada implementasi Secure Handshake Protocol sebagai mekanisme kontrol utama.Alih-alih bertindak sebagai perantara yang menyimpan kunci akses, ZENITH memanfaatkan Auth0 untuk membangun lapisan Token Vault yang terisolasi. Dalam skenario praktis, misalnya ketika pengguna meminta sistem untuk membuat dan mempublikasikan konten ke repositori GitHub, proses tidak langsung dieksekusi. Sebaliknya, sistem memicu serangkaian secure handshake sebagai berikut:<ol> <li style="text-align: justify">Pengguna memverifikasi identitas melalui Multi-Factor Authentication (MFA).</li> <li style="text-align: justify">Permintaan dievaluasi menggunakan kebijakan akses granular (fine-grained policies) berbasis OpenFGA.</li> <li style="text-align: justify">Token Vault hanya menerbitkan scoped token berdurasi pendek yang spesifik untuk aksi tersebut.</li> </ol>Salah satu komponen paling kritis dalam pendekatan ini adalah implementasi MFA step-up secara real-time menggunakan Google Authenticator. Mekanisme ini bukan sekadar fitur tambahan, melainkan titik kontrol di mana “aksi agentik” benar-benar berada di bawah verifikasi manusia. Sebelum ZENITH mengeksekusi operasi sensitif, pengguna harus membuktikan kehadirannya secara real-time, melampaui sekadar validitas sesi. Dengan pendekatan ini, large language model (LLM) tidak pernah memiliki akses langsung terhadap kredensial asli pengguna. Dengan memanfaatkan protokol Universal Login dan OpenID Connect (OIDC) dari Auth0, ZENITH tidak pernah mengakses raw secrets. Sebagai gantinya, sistem hanya menerima Action Token yang bersifat sementara dan terikat secara kriptografis pada identitas terverifikasi. Setiap “Neural Request”, baik itu sintesis gambar resolusi tinggi maupun produksi video sinematik, memiliki keterkaitan yang jelas dengan identitas pengguna yang sah. Integrasi Auth0 dalam konteks ini melampaui sekadar fitur autentikasi. Ia berfungsi sebagai sebuah Governance Layer, di mana identitas menjadi perimeter utama dalam arsitektur sistem. Tanpa autentikasi melalui Token Vault, agen-agen AI seperti Copywriter, Illustrator, dan Animator secara sistemik tidak dapat diinisialisasi.Pendekatan dekopling (pemisahan) antara agensi AI dan identitas ini memungkinkan fokus pengembangan diarahkan pada kompleksitas orkestrasi model multi-modal, tanpa mengorbankan aspek keamanan. Ketika Auth0 menangani manajemen sesi dan pertukaran token secara aman, ZENITH memiliki ruang untuk mengeksplorasi kapabilitas model terbaru seperti Gemini 3.1 Pro secara optimal. Proyek ini menunjukkan bahwa generasi AI berikutnya tidak hanya ditentukan oleh kualitas prompt engineering, tetapi juga oleh kekuatan infrastruktur keamanannya. Melalui integrasi Auth0 Token Vault dan OpenFGA, ZENITH bertransformasi dari sekadar demonstrasi teknologi menjadi sebuah Authorized Agent yang siap untuk diimplementasikan pada lingkungan produksi.Kita sedang memasuki era di mana AI tidak hanya menghasilkan rekomendasi, tetapi juga melakukan tindakan nyata, baik di ranah digital maupun fisik. Dalam konteks tersebut, kepercayaan (trust) menjadi mata uang utama. Auth0 menyediakan fondasi keamanannya, sementara ZENITH menghadirkan visinya. Keduanya merepresentasikan cetak biru bagi sistem AI yang tidak hanya canggih, tetapi juga aman dan tetap berpusat pada manusia.References Mensah, Frank. (2024). Zero Trust Architecture: A Comprehensive Review of Principles, Implementation Strategies, and Future Directions in Enterprise Cybersecurity. International Journal of Academic and Industrial Research Innovations (IJAIRI). 10. 339-346 Ejiofor, Oluomachi &amp; Olusoga, Oluwafemi &amp; Akinsola, Ahmed. (2025). Zero trust architecture: A paradigm shift in network security. Computer Science &amp; IT Research Journal. 6. 104-124. 10.51594/csitrj.v6i3.1871 Ume Nisa, Muhammad Shirazi, Mohamed Ali Saip, Muhammad Syafiq Mohd Pozi, Agentic AI: The age of reasoning. A review, Journal of Automation and Intelligence, Volume 5, Issue 1, 2026, Pages 69-89, ISSN 2949-8554, https://doi.org/10.1016/j.jai.2025.08.003 Zhou, Qian &amp; Elbadry, Mohammed &amp; ye, Fan &amp; Yang, Yuanyuan. (2020). Towards Fine-Grained Access Control in Enterprise-Scale Internet-of-Things. IEEE Transactions on Mobile Computing. PP. 1-1. 10.1109/TMC.2020.2984700 Wissuchek, Christopher &amp; Zschech, Patrick. (2025). Challenges in Managing the Relationship Between Agentic AI Systems and Humans in Organizations. 10.1007/978-3-031-94193-1_1 Khalid, Omer &amp; Farooqi, Ammad &amp; Bilal, Muhammad. (2025). Agentic AI: A Review, Applications, and Open Research Challenges. 10.20944/preprints202512.0592.v1 Ranjan Sapkota, Konstantinos I. Roumeliotis, Manoj Karkee, AI Agents vs. Agentic AI: A Conceptual taxonomy, applications and challenges, Information Fusion, Volume 126, Part B, 2026, 103599, ISSN 1566-2535, https://doi.org/10.1016/j.inffus.2025.103599 Wu, Yuhao &amp; Yang, Ke &amp; Roesner, Franziska &amp; Kohno, Tadayoshi &amp; Zhang, Ning &amp; Iqbal, Umar. (2025). Towards Automating Data Access Permissions in AI Agents. 10.48550/arXiv.2511.17959 Batool, Amna &amp; Zowghi, Didar &amp; Bano, Muneera. (2024). AI Governance: A Systematic Literature Review. 10.21203/rs.3.rs-4784792/v1 Akhmad, K. (2026). Zenith: An Agentic Governance Framework with Sovereign AI and Fine-Grained Authorization (V1.0.0). Zenodo. https://doi.org/10.5281/zenodo.19467804

We often still think of artificial intelligence (AI) as nothing more than a chatbot—a place to ask questions, find answers, or simply have a conversation. But what if AI could go beyond that role and act as a personal assistant that actually performs complex tasks, from pushing code to a GitHub repository to managing schedules using Google Calendar, without compromising user data security? This question forms the foundation of ZENITH: a next-generation AI orchestrator that focuses not only on intelligence but also prioritizes identity sovereignty as a core principle in every action it performs.Building a complex AI system like ZENITH, capable of simultaneously coordinating advanced models such as Google Gemini 3.1 Pro, Imagen, and Veo 3.1, presents a fundamental paradox: how to design a system powerful enough to “act” on behalf of the user, yet secure enough that it never touches the user’s most sensitive credentials? In AI agent development, a common mistake still frequently encountered is the use of hardcoded API keys or storing credentials in frontend state—both of which are vulnerable to exploitation. For ZENITH, such an approach is not an option. Therefore, a Zero-Trust architecture is implemented comprehensively, with Auth0 Token Vault serving as the primary foundation for identity and access management. Empirical experience shows that exposed credentials can be exploited in a matter of seconds. Automated bots are capable of massively consuming API quotas, underscoring that in modern digital ecosystems, unauthorized access is not merely a possibility but an inevitability. To address this challenge, ZENITH’s development focuses on implementing the Secure Handshake Protocol as the primary control mechanism.Instead of acting as an intermediary that stores access keys, ZENITH leverages Auth0 to build an isolated Token Vault layer. In a practical scenario, for example, when a user requests the system to create and publish content to a GitHub repository, the process is not executed directly. Instead, the system triggers a series of secure handshakes as follows:<ol> <li style="text-align: justify">The user verifies their identity via Multi-Factor Authentication (MFA).</li> <li style="text-align: justify">The request is evaluated using fine-grained access policies based on OpenFGA.</li> <li style="text-align: justify">The Token Vault issues only short-lived, scoped tokens specific to that action.</li> </ol>One of the most critical components of this approach is the real-time implementation of step-up MFA using Google Authenticator. This mechanism is not merely an additional feature, but a control point where “agent actions” are truly subject to human verification. Before ZENITH executes sensitive operations, the user must prove their presence in real-time, going beyond mere session validity. With this approach, the large language model (LLM) never has direct access to the user’s original credentials. By leveraging Auth0’s Universal Login and OpenID Connect (OIDC) protocols, ZENITH never accesses raw secrets. Instead, the system only receives temporary Action Tokens cryptographically bound to a verified identity. Every “Neural Request”—whether it’s high-resolution image synthesis or cinematic video production—has a clear link to the user’s legitimate identity. Auth0 integration in this context goes beyond mere authentication features. It functions as a Governance Layer, where identity serves as the primary perimeter within the system architecture. Without authentication via the Token Vault, AI agents such as Copywriter, Illustrator, and Animator cannot be systematically initialized.This decoupling approach between AI agents and identity allows development focus to be directed toward the complexity of multi-modal model orchestration, without compromising security. While Auth0 handles session management and secure token exchange, ZENITH has the freedom to optimally explore the capabilities of the latest models, such as Gemini 3.1 Pro. This project demonstrates that the next generation of AI is not only defined by the quality of prompt engineering but also by the strength of its security infrastructure. Through the integration of Auth0 Token Vault and OpenFGA, ZENITH has transformed from a mere technology demonstration into an Authorized Agent ready for deployment in a production environment.We are entering an era where AI not only generates recommendations but also takes tangible actions, both in the digital and physical realms. In this context, trust becomes the primary currency. Auth0 provides the security foundation, while ZENITH brings its vision to life. Together, they represent a blueprint for AI systems that are not only advanced but also secure and human-centered.References Mensah, Frank. (2024). Zero Trust Architecture: A Comprehensive Review of Principles, Implementation Strategies, and Future Directions in Enterprise Cybersecurity. International Journal of Academic and Industrial Research Innovations (IJAIRI). 10. 339-346 Ejiofor, Oluomachi &amp; Olusoga, Oluwafemi &amp; Akinsola, Ahmed. (2025). Zero Trust Architecture: A Paradigm Shift in Network Security. Computer Science &amp; IT Research Journal. 6. 104–124. 10.51594/csitrj.v6i3.1871 Ume Nisa, Muhammad Shirazi, Mohamed Ali Saip, Muhammad Syafiq Mohd Pozi, Agentic AI: The Age of Reasoning. A review, Journal of Automation and Intelligence, Volume 5, Issue 1, 2026, Pages 69–89, ISSN 2949-8554, https://doi.org/10.1016/j.jai.2025.08.003 Zhou, Qian &amp; Elbadry, Mohammed &amp; Ye, Fan &amp; Yang, Yuanyuan. (2020). Towards Fine-Grained Access Control in Enterprise-Scale Internet-of-Things. IEEE Transactions on Mobile Computing. pp. 1–1. 10.1109/TMC.2020.2984700 Wissuchek, Christopher &amp; Zschech, Patrick. (2025). Challenges in Managing the Relationship Between Agentic AI Systems and Humans in Organizations. 10.1007/978-3-031-94193-1_1 Khalid, Omer &amp; Farooqi, Ammad &amp; Bilal, Muhammad. (2025). Agentic AI: A Review, Applications, and Open Research Challenges. 10.20944/preprints202512.0592.v1 Ranjan Sapkota, Konstantinos I. Roumeliotis, Manoj Karkee, AI Agents vs. Agentic AI: A Conceptual Taxonomy, Applications, and Challenges, Information Fusion, Volume 126, Part B, 2026, 103599, ISSN 1566-2535, https://doi.org/10.1016/j.inffus.2025.103599 Wu, Yuhao &amp; Yang, Ke &amp; Roesner, Franziska &amp; Kohno, Tadayoshi &amp; Zhang, Ning &amp; Iqbal, Umar. (2025). Towards Automating Data Access Permissions in AI Agents. 10.48550/arXiv.2511.17959 Batool, Amna &amp; Zowghi, Didar &amp; Bano, Muneera. (2024). AI Governance: A Systematic Literature Review. 10.21203/rs.3.rs-4784792/v1 Akhmad, K. (2026). Zenith: An Agentic Governance Framework with Sovereign AI and Fine-Grained Authorization (V1.0.0). Zenodo. https://doi.org/10.5281/zenodo.19467804

Dekopling Agensi dan Identitas: Orkestrasi AI Berbasis Arsitektur Zero-Trust

Decoupling Agency and Identity: AI Orchestration Based on a Zero-Trust Architecture

The rapid evolution of Internet of Things (IoT) systems, smart cities, and intelligent infrastructures is reshaping networking and cybersecurity requirements. Traditional architectures struggle to meet the demands of scalability, adaptability, and security in dynamic environments. This paper reviews recent advances in AI-driven networking, edge intelligence, and modern cybersecurity paradigms, including zero-trust and AI-based threat detection. It also examines proactive congestion control mechanisms for maintaining Quality of Service (QoS). By integrating artificial intelligence with distributed computing and adaptive security frameworks, next-generation networks can achieve improved reliability, efficiency, and resilience.

1. IntroductionThe increasing deployment of IoT devices and real-time applications has introduced significant challenges in modern communication networks. Smart city services, intelligent transportation systems, and critical infrastructures require low latency, high reliability, and strong security guarantees. Traditional approaches based on static routing and perimeter-based security are insufficient in such dynamic and heterogeneous environments [1], [2]. Recent developments are driving a transition toward intelligent, decentralized network architectures, where Artificial Intelligence (AI), edge computing, and adaptive security mechanisms operate jointly. These systems enable real-time decision-making, predictive analysis, and automated response to network conditions.2. AI-Driven Networking2.1 Intelligent RoutingAI-based routing leverages reinforcement learning (RL) and deep learning to dynamically select forwarding paths based on observed network conditions. A typical RL-based routing model defines:• State: link quality, queue length, delay, packet loss• Action: next-hop selection• Reward: function of latency, delivery ratio, and congestionSuch formulations enable continuous learning and adaptation to changing traffic patterns. Recent studies demonstrate that Q-learning-based next-hop selection improves packet delivery ratio and reduces end-to-end delay in dynamic IoT environments [11].2.2 Predictive Network ManagementAI-driven predictive models analyze temporal traffic patterns to anticipate congestion and link degradation. For example, time-series and deep learning models can forecast traffic load at edge nodes, enabling preemptive traffic rerouting and load balancing [3]. This shifts network control from reactive to proactive operation.3. Edge and Fog ComputingEdge and fog computing extend processing capabilities closer to data sources, reducing reliance on centralized cloud infrastructures. Key advantages include Reduced end-to-end latency, Localized processing of time-sensitive data, Reduced backbone traffic, and Enhanced privacy through local enforcement. The integration of Edge AI allows inference tasks (e.g., anomaly detection, traffic prediction) to be executed directly at gateways or edge servers, enabling near real-time decision-making [4], [5]. Figure 1: Multi-layer architecture illustrating IoT devices, edge gateways, fog nodes, and cloud platforms supporting real-time processing and secure communication.The architecture in Figure 1 illustrates hierarchical data processing: sensor data is first handled at the edge (filtering, aggregation), then processed at fog nodes (analytics, coordination), and finally stored or globally optimized in the cloud. This distribution reduces latency and improves scalability.4. Cybersecurity in Modern Networks4.1 Zero-Trust ArchitectureZero-Trust Architecture (ZTA) enforces continuous verification of users and devices, independent of network location. Access decisions are based on identity, device posture, and contextual policies, reducing the attack surface in distributed systems [6].4.2 AI-Based Threat DetectionMachine learning models can identify anomalous traffic patterns by analyzing features such as flow statistics, packet timing, and protocol behavior. These systems enable detection of zero-day attacks. identification of lateral movement, and automated incident response.Recent approaches combine deep learning with edge deployment to achieve low-latency threat detection [7].4.3 Emerging Security ParadigmsEmerging approaches such as semantic communications aim to transmit only task-relevant information rather than raw data, improving both efficiency and security [8]. Additionally, blockchain-based mechanisms are being explored for decentralized identity and trust management.5. Congestion Control and Traffic EngineeringEfficient traffic management is essential for maintaining QoS in large-scale networks.Traditional congestion control mechanisms rely on packet loss or delay as implicit signals, leading to delayed response. In contrast, modern approaches emphasize explicit congestion signaling, queue-aware routing decisions, and cross-layer optimization. For example, congestion-aware routing can incorporate queue length or buffer occupancy into routing decisions. Proactive mechanisms, such as early congestion notification, allow upstream nodes to adjust transmission rates before packet loss occurs [10], [11]. When combined with AI-based prediction models, these mechanisms enable anticipatory congestion control, improving throughput stability and reducing latency variability [9].6. Integration of Emerging TechnologiesTable 1: Key Technologies in Future Networking and Cybersecurity<table> <tbody> <tr> <td>Technology</td> <td>Networking Role</td> <td>Security Contribution</td> </tr> <tr> <td>AI / ML / LLMs</td> <td>Adaptive Routing, Prediction</td> <td>Intelligent anomaly detection</td> </tr> <tr> <td>Edge AI</td> <td>Real-time local processing</td> <td>Low-latency threat detection</td> </tr> <tr> <td>Zero-Trust</td> <td>Access Control</td> <td>Continuous Authentication</td> </tr> <tr> <td>Blockchain</td> <td>DIstributed Coordination</td> <td>Data Integrity</td> </tr> <tr> <td>AI-based Congestion Control</td> <td>Traffic Optimization</td> <td>Mitigates overload conditions</td> </tr> </tbody> </table>The integration of these technologies presented in Table 1 enables coordinated operation across networking and security layers. For example, edge-based anomaly detection can inform routing decisions, while congestion signals can be used to prioritize critical traffic.7. Challenges and Future DirectionsDespite recent progress, several challenges remain:<ul> <li style="text-align: justify">Scalability: AI models must operate efficiently across thousands of distributed nodes</li> <li style="text-align: justify">Resource constraints: IoT devices have limited computation and energy capacity</li> <li style="text-align: justify">Data privacy: centralized learning raises privacy concerns</li> <li style="text-align: justify">Interoperability: integration across heterogeneous systems remains complexPromising research directions include lightweight and distributed AI models, federated learning for privacy preservation, self-healing and autonomous networks, joint optimization of routing, congestion control, and security.</li> </ul>8. ConclusionThe convergence of AI-driven networking, edge intelligence, and adaptive cybersecurity frameworks is shaping the future of communication systems. Intelligent routing, predictive management, and proactive congestion control provide a foundation for scalable and resilient networks. These advancements are critical for supporting the growing demands of smart systems and ensuring secure and efficient network operation.References[1] H. Du et al., “The age of generative AI and AI-generated everything,” IEEE Netw., vol. 38, no. 6, pp. 501–512, Nov./Dec. 2024, doi: 10.1109/MNET.2024.3422241.[2] Y. Huang et al., “Large language models for networking: Applications, enabling techniques, and challenges,” IEEE Netw., vol. 39, no. 1, pp. 235–242, Jan./Feb. 2024.[3] R. Sun et al., “Knowledge-driven deep learning paradigms for wireless network optimization in 6G,” IEEE Netw., vol. 38, no. 2, pp. 70–78, Mar./Apr. 2024.[4] S. S. Gill et al., “Edge AI: A taxonomy, systematic review and future directions,” Cluster Comput., vol. 28, Art. no. 18, 2025.[5] C. Wang et al., “The security and privacy of mobile edge computing: An artificial intelligence perspective,” arXiv preprint arXiv:2401.01589, 2024.[6] NIST, Zero Trust Architecture, SP 800-207, Gaithersburg, MD, USA, Aug. 2020.Networks, IEEE Std 802.1Q-2024, 2024.[7] Y. Wang et al., “Advanced persistent threat detection,” IEEE Netw., 2024.[8] Z. Yang et al., “Secure semantic communications,” IEEE Netw., 2024.[9] H. Zhang et al., “Deep learning-based traffic prediction and congestion control for smart cities,” IEEE Netw., vol. 35, no. 2, pp. 198–204, Mar./Apr. 2021.Malaysia, 2010.[10] M. M. Kadhum, “Enhancing the performance and reliability of IoT systems using predictive route recovery mechanism,” in Proc. IEEE TENCON, Dec. 2024, pp. 489–493.[11] M. M. Kadhum, “Machine learning technique for route discovery in dynamic Internet of Things environments: A theoretical perspective,” in Proc. Int. Conf. Sens. Technol. (ICST), Dec. 2024, pp. 1–4.

AI-Driven Networking and Cybersecurity for Next-Generation Smart Systems

As network environments grow in scale, heterogeneity, and dynamism, traditional signature‑based intrusion detection struggles to keep pace. Unsupervised learning is a technique that detects anomalies without requiring labeled attack examples has become central to modern Intrusion Detection System (IDS) design because it can discover novel and polymorphic threats in complex trafficModern network traffics carry diverse, high‑velocity flows: encrypted traffic (TLS 1.3, QUIC), ephemeral micro-services communications (service meshes, gRPC), IoT telemetry, and cloud‑native east‑west traffic. These characteristics complicate feature extraction and baseline modeling in several ways, i.e.: High dimensionality and heterogeneity; Concept drift; Low event signal-to-noise ratio, where malicious events are sparse and can be camouflaged by benign but unusual behavior. Unsupervised IDS must therefore build robust, adaptive representations that tolerate benign variability while remaining sensitive to subtle anomalies.On the other hand, modern attackers leverage techniques that defeat naïve anomaly detectors such as: (i) Living‑off‑the‑land and file-less attacks by abusing legitimate processes and APIs, producing activity that superficially resembles benign behavior. (ii) Slow‑burn or low‑and‑slow exfiltration by spreading malicious actions across long time windows to avoid short‑term thresholds. (iii) Polymorphism and adversarial perturbation by modifying payloads or timings to remain within learned normal bounds or to exploit detector weaknesses, and (iv) Multi‑stage campaigns, including: reconnaissance, lateral movement, and data staging each exhibit different signatures, requiring cross‑stage correlation.Detecting such threats requires unsupervised systems that can (1) model temporal dependencies across multiple scales, (2) learn semantically meaningful representations of flows (beyond raw packet counts), and (3) incorporate context (host, user, service) to reduce false positives. Ensemble and hierarchical approaches that combine short‑window detectors with longer‑term models are often effective (Mirsky, et al. 2018).Unsupervised methods for network anomaly detection comprise various approaches that learn normal behavior without needing large sets of labeled attack examples. A typical pipeline (see Figure 1) first converts raw network data into features; an unsupervised model then captures patterns from mostly benign traffic; departures from those patterns are scored; and a thresholding stage turns scores into alerts. Methods within this framework differ in how they define “normal” (e.g., distributional, geometric, predictive, or reconstruction-based), how they compute anomaly scores, and how they preserve stability in the face of noise, multi‑modality, and non-stationarity (Wang, et al. 2021; Gamage, et al. 2020).<ul> <li style="text-align: justify">Memory‑prediction architectures, i.e.: models that incorporate explicit memory components and predictive objectives are gaining traction in unsupervised IDS for several reasons: Predictive power: Training models to predict future observations (next flow features, sequence of events) encourages learning of temporal regularities; deviations between predictions and actual observations highlight anomalies.</li> <li style="text-align: justify">Long‑term context: External memory mechanisms (neural memory, attention, or differentiable memory modules) allow models to recall and compare current behavior with historical patterns across hosts, users, or services.</li> <li style="text-align: justify">Fast adaptation: Memory components facilitate few‑shot assimilation of new benign patterns without catastrophic forgetting, aiding concept drift handling</li> </ul> <img src="/uploads/content/1778081367251-222489688.webp">Figure 1. Conceptual Framework of Unsupervised Anomaly Detection with On-the-Fly and Memory-Aware Learning (Alqithami, et al. 2025)Thus, unsupervised learning is vital for contemporary IDS, as it enables detection of new and complex threats in evolving network settings. Advances depend on strong representation learning, time-aware and memory‑enhanced predictive models, and thoughtful operational deployment. Going forward, developments will likely unite scalable unsupervised models, continual learning to handle concept drift, and feedback mechanisms for analysts that link detection to response.ReferenceMirsky, Yisroel, Tomer Doitshman, Yuval Elovici, and Asaf Shabtai. 2018. “Kitsune: An Ensemble of Autoencoders for Online Network Intrusion Detection.” In Proceedings of the Network and Distributed System Security Symposium (NDSS) 2018.Saad Alqithami, Deris Stiawan, Rahmat Budiarto, 2025, On-the-fly, memory-aware unsupervised learning for network anomaly detection: A systematic literature review, e-Prime - Advances in Electrical Engineering, Electronics and Energy, 101146, doi:10.1016/j.prime.2025.101146.S. Wang, J.F. Balarezo, S. Kandeepan, A. Al-Hourani, K. Gome. Chavez, B. Ru-binstein, Machine learning in network anomaly detection: A survey, IEEE Access 9 (2021) 152904–152922, http://dx.doi.org/10.1109/ACCESS.2021.3125219.S. Gamage, J. Samarabandu, Deep learning methods in network intrusion detection: A survey and an objective comparison, J. Netw. Comput. Appl. 169 (2020) 102767, http://dx.doi.org/10.1016/j.jnca.2020.102767.

IDS Using Unsupervised Learning: Adapting to Modern Network Traffic, Sophisticated Attacks, and Memory‑Prediction Approaches

Perkembangan teknologi digital telah menghasilkan pertumbuhan data tidak terstruktur dalam jumlah yang sangat besar, termasuk laporan keamanan siber, media sosial, dokumen digital, dan berbagai bentuk komunikasi elektronik lainnya. Kondisi tersebut mendorong kebutuhan terhadap teknologi yang mampu memahami dan mengolah data teks secara otomatis. Natural Language Processing (NLP) berkembang sebagai salah satu pendekatan penting dalam mendukung proses ekstraksi informasi, klasifikasi teks, analisis konteks, serta pengembangan sistem cerdas berbasis data. Pada bidang keamanan siber, NLP dimanfaatkan untuk mendukung Cyber Threat Intelligence melalui proses ekstraksi informasi ancaman dari laporan keamanan siber yang tidak terstruktur. Informasi seperti malware, Indicators of Compromise (IoC), teknik serangan, dan pola aktivitas aktor ancaman dapat diidentifikasi secara otomatis untuk menghasilkan intelligence yang lebih terstruktur dan mudah dianalisis. Selain itu, perkembangan transformer architecture dan Large Language Models (LLM) semakin memperluas kemampuan NLP dalam memahami konteks bahasa secara lebih kompleks dan adaptif. Integrasi antara NLP, artificial intelligence, deep learning, dan cybersecurity memberikan peluang besar dalam membangun sistem analisis ancaman siber yang lebih cerdas, otomatis, dan kontekstual. Penelitian ini membahas perkembangan NLP, penerapannya pada cybersecurity dan cyber threat intelligence, tantangan pengolahan Bahasa Indonesia, serta arah pengembangan masa depan berbasis Large Language Models untuk mendukung transformasi digital dan penguatan keamanan siber.

Advances in digital technology have led to a massive increase in unstructured data, including cybersecurity reports, social media posts, digital documents, and various other forms of electronic communication. This has driven the need for technologies capable of automatically understanding and processing textual data. Natural Language Processing (NLP) has emerged as a key approach in supporting information extraction, text classification, contextual analysis, and the development of data-driven intelligent systems. In the field of cybersecurity, NLP is utilized to support Cyber Threat Intelligence through the process of extracting threat information from unstructured cybersecurity reports. Information such as malware, Indicators of Compromise (IoC), attack techniques, and patterns of threat actor activity can be automatically identified to generate more structured and easily analyzable intelligence. Furthermore, advancements in transformer architecture and Large Language Models (LLMs) are expanding NLP’s capabilities to understand language context in a more complex and adaptive manner. The integration of NLP, artificial intelligence, deep learning, and cybersecurity presents significant opportunities for building smarter, automated, and context-aware cyber threat analysis systems. This research discusses the development of NLP, its application in cybersecurity and cyber threat intelligence, the challenges of processing the Indonesian language, and future development directions based on Large Language Models to support digital transformation and strengthen cybersecurity.

Menurut Xue and Liu [1], perkembangan teknologi digital dalam beberapa tahun terakhir telah menghasilkan pertumbuhan data tidak terstruktur dalam jumlah yang sangat besar, seperti dokumen digital, media sosial, laporan keamanan siber, artikel daring, dan komunikasi elektronik lainnya. Ferrag et al. [2] menjelaskan bahwa kebutuhan terhadap teknologi yang mampu memahami dan mengolah data teks secara otomatis menjadi semakin penting dalam mendukung pengambilan keputusan berbasis data dan sistem cerdas modern. Selain itu, perkembangan transformer architecture dan Large Language Models (LLM) juga semakin memperluas kemampuan NLP dalam memahami konteks bahasa manusia secara lebih kompleks dan adaptif.Arazzi et al. [3] menjelaskan bahwa Natural Language Processing berkembang sebagai solusi untuk membantu proses ekstraksi informasi, klasifikasi teks, analisis konteks, dan pembentukan sistem cerdas yang mampu memahami bahasa alami manusia secara otomatis. NLP mengintegrasikan berbagai disiplin ilmu seperti linguistik, machine learning, deep learning, data mining, dan artificial intelligence untuk menghasilkan pendekatan yang mampu mengubah data teks menjadi informasi yang lebih terstruktur dan mudah dianalisis. Dengan kemampuan tersebut, NLP saat ini banyak diterapkan pada berbagai bidang seperti cybersecurity, business intelligence, analisis media sosial, healthcare, hingga sistem pendukung keputusan.Dalam bidang keamanan siber, penelitian diarahkan pada penerapan Cyber Threat Intelligence untuk mendukung proses deteksi, mitigasi, dan analisis ancaman digital secara otomatis. Menurut Ismail [4] sebagian besar informasi ancaman siber masih tersedia dalam bentuk laporan teks tidak terstruktur seperti threat intelligence reports, incident reports, security advisories, dan malware analysis reports. Kondisi tersebut menyebabkan proses analisis ancaman siber sering kali membutuhkan waktu yang lama apabila dilakukan secara manual. Oleh karena itu, NLP dimanfaatkan untuk membantu proses ekstraksi informasi ancaman siber sehingga menghasilkan data intelligence yang lebih terstruktur dan mudah dianalisis.Penelitian pada bidang ini difokuskan pada pengembangan metode text mining dan information extraction untuk mengidentifikasi berbagai informasi penting dari laporan ancaman siber, seperti nama malware, indikator kompromi (Indicators of Compromise/IoC), teknik serangan, target serangan, serta pola aktivitas yang dilakukan oleh aktor ancaman. Menurut penelitian yang dilakukan oleh Albarrak et al. [5] pendekatan NLP pada cyber threat intelligence mampu meningkatkan efektivitas proses identifikasi ancaman serta membantu pengembangan sistem keamanan siber yang lebih proaktif dan adaptif. Selain pada domain cybersecurity, penelitian juga diarahkan pada analisis malware Android. Raju et al. [6] menjelaskan bahwa perkembangan perangkat mobile yang sangat pesat menyebabkan sistem operasi Android menjadi salah satu target utama serangan malware. Malware Android terus berkembang dengan berbagai teknik serangan yang semakin kompleks sehingga membutuhkan pendekatan analisis yang lebih adaptif dan otomatis. Dalam konteks ini, NLP digunakan untuk membantu proses analisis laporan ancaman siber terkait malware Android guna menghasilkan dataset intelligence yang dapat digunakan dalam pengembangan model deteksi malware berbasis machine learning dan deep learning.Pengembangan dataset berbasis threat intelligence menjadi salah satu fokus penting dalam penelitian. Menurut penelitian Rahman et al. [7] dan Xu et al. [8] integrasi antara NLP, artificial intelligence, dan cybersecurity memberikan peluang besar dalam membangun sistem keamanan siber yang lebih cerdas dan otomatis. Dataset yang dihasilkan melalui proses ekstraksi otomatis dari laporan keamanan siber diharapkan mampu memberikan konteks intelligence yang lebih kaya dibandingkan dataset malware konvensional. Data yang telah diekstraksi kemudian digunakan dalam proses klasifikasi, prediksi, maupun analisis pola ancaman menggunakan berbagai pendekatan artificial intelligence dan deep learning.Dalam implementasinya, penelitian juga memanfaatkan berbagai pendekatan machine learning dan deep learning untuk mendukung proses analisis bahasa alami. Ainslie et al. [9] menjelaskan bahwa teknik seperti text classification, topic modeling, sentiment analysis, named entity recognition (NER), dan relation extraction memiliki peran penting dalam memahami struktur dan konteks data teks secara lebih mendalam. Selain itu, perkembangan transformer-based language models dan Large Language Models membuka peluang baru dalam pengembangan sistem analisis ancaman siber berbasis artificial intelligence yang mampu melakukan analisis secara otomatis dan real-time.Pemanfaatan NLP juga diterapkan pada berbagai bidang lain seperti business intelligence, analisis media sosial, data analytics, dan sistem pendukung keputusan. Menurut penelitian Ismail [4] NLP dapat digunakan untuk menganalisis opini pelanggan, memahami tren pasar, serta mengidentifikasi pola perilaku pengguna dari data digital yang tersedia. Pada analisis media sosial, NLP dimanfaatkan untuk melakukan sentiment analysis, deteksi topik, dan pemetaan opini publik terhadap isu tertentu. Hal tersebut menunjukkan bahwa NLP memiliki peran penting dalam mendukung transformasi digital berbasis data pada berbagai sektor industri dan organisasi modern.Pengembangan penelitian pada bidang NLP juga diarahkan untuk mendukung pengolahan Bahasa Indonesia yang masih memiliki berbagai tantangan dibandingkan bahasa dengan sumber daya tinggi seperti Bahasa Inggris. Tantangan tersebut meliputi keterbatasan dataset, variasi struktur bahasa, penggunaan bahasa tidak formal, serta minimnya sumber daya linguistik yang tersedia. Oleh karena itu, penelitian pada bidang NLP diharapkan dapat memberikan kontribusi dalam pengembangan teknologi pengolahan bahasa alami yang lebih adaptif terhadap karakteristik Bahasa Indonesia, khususnya pada domain keamanan siber dan analisis intelligence.Masa depan, arah pengembangan kepakaran difokuskan pada integrasi antara NLP, artificial intelligence, cyber intelligence, dan Large Language Models untuk membangun sistem analisis ancaman siber yang lebih cerdas, otomatis, dan kontekstual. Menurut Ahi and Valizadeh [10], pemanfaatan LLM pada bidang cybersecurity memiliki potensi besar dalam meningkatkan kemampuan sistem dalam memahami konteks ancaman, melakukan reasoning terhadap data ancaman siber, serta menghasilkan informasi intelligence yang lebih akurat dan relevan. Dengan perkembangan teknologi tersebut, NLP diharapkan mampu menjadi salah satu fondasi penting dalam mendukung transformasi digital, penguatan keamanan siber, dan pengembangan sistem cerdas berbasis data di masa mendatang.Referensi [1] H. Xue and W. Liu, “Bibliometric Analysis of Natural Language Processing Technology in Education: Hot Topics, Frontier Evolution, and Future Prospects,” SAGE Open, vol. 15, no. 1, 2025, doi: 10.1177/21582440251319891. [2] M. A. Ferrag et al., “Revolutionizing Cyber Threat Detection with Large Language Models: A Privacy-Preserving BERT-Based Lightweight Model for IoT/IIoT Devices,” IEEE Access, vol. 12, pp. 23733–23750, 2024, doi: 10.1109/ACCESS.2024.3363469. [3] M. Arazzi et al., “NLP-based techniques for Cyber Threat Intelligence,” Computer Science Review, vol. 58, p. 100765, 2025, doi: 10.1016/j.cosrev.2025.100765. [4] W. S. Ismail, “Threat Detection and Response Using AI and NLP in Cybersecurity,” Journal of Internet Services and Information Security, vol. 14, no. 1, pp. 195–205, 2024, doi: 10.58346/JISIS.2024.I1.013. [5] M. Albarrak, K. Salonitis, and S. Jagtap, “Natural Language Processing (NLP)-Based Frameworks for Cyber Threat Intelligence and Early Prediction of Cyberattacks in Industry 4.0: A Systematic Literature Review,” Applied Sciences, vol. 16, no. 2, p. 619, 2026, doi: 10.3390/app16020619. [6] A. D. Raju, I. Y. Abualhaol, R. S. Giagone, Y. Zhou, and S. Huang, “A Survey on Cross-Architectural IoT Malware Threat Hunting,” IEEE Access, vol. 9, pp. 91686–91708, 2021, doi: 10.1109/ACCESS.2021.3091427. [7] M. A. Rahman et al., “A Survey of Large Language Models (LLMs) for Cybersecurity: Opportunities and Directions,” in Proc. 2025 IEEE Int. Conf. Big Data (BigData), 2025, pp. 4333–4342, doi: 10.1109/BigData66926.2025.11402639. [8] H. Xu et al., “Large Language Models for Cyber Security: A Systematic Literature Review,” ACM Transactions on Software Engineering and Methodology, 2025, doi: 10.1145/3769676. [9] S. Ainslie, D. Thompson, S. Maynard, and A. Ahmad, “Cyber-threat intelligence for security decision-making: A review and research agenda for practice,” Computers and Security, vol. 132, 2023, doi: 10.1016/j.cose.2023.103352. [10] K. Ahi and S. Valizadeh, “Large Language Models (LLMs) and Generative AI in Cybersecurity and Privacy: A Survey of Dual-Use Risks, AI-Generated Malware, Explainability, and Defensive Strategies,” in 2025 Silicon Valley Cybersecurity Conference (SVCC), 2025, pp. 1–8, doi: 10.1109/SVCC65277.2025.11133642.

According to Xue and Liu [1], the development of digital technology in recent years has led to a massive increase in unstructured data, such as digital documents, social media, cybersecurity reports, online articles, and other forms of electronic communication. Ferrag et al. [2] explain that the need for technologies capable of automatically understanding and processing text data has become increasingly important in supporting data-driven decision-making and modern intelligent systems. Furthermore, advancements in transformer architecture and Large Language Models (LLMs) have further expanded NLP’s capabilities to understand human language context in a more complex and adaptive manner.Arazzi et al. [3] explain that Natural Language Processing has evolved as a solution to assist in information extraction, text classification, context analysis, and the development of intelligent systems capable of automatically understanding human natural language. NLP integrates various disciplines such as linguistics, machine learning, deep learning, data mining, and artificial intelligence to produce approaches capable of transforming text data into more structured and easily analyzable information. With these capabilities, NLP is currently widely applied in various fields such as cybersecurity, business intelligence, social media analysis, healthcare, and decision support systems.In the field of cybersecurity, research is focused on the application of Cyber Threat Intelligence to support the automated detection, mitigation, and analysis of digital threats. According to Ismail [4], most cyber threat information is still available in the form of unstructured text reports such as threat intelligence reports, incident reports, security advisories, and malware analysis reports. This situation often results in the cyber threat analysis process taking a long time when performed manually. Therefore, NLP is utilized to assist in the extraction of cyber threat information, thereby generating more structured and easily analyzable intelligence data.Research in this field focuses on developing text mining and information extraction methods to identify various critical details from cyber threat reports, such as malware names, indicators of compromise (IoC), attack techniques, attack targets, and activity patterns carried out by threat actors. According to research conducted by Albarrak et al. [5], the application of NLP in cyber threat intelligence enhances the effectiveness of threat identification processes and aids in the development of more proactive and adaptive cybersecurity systems. Beyond the cybersecurity domain, research is also directed toward the analysis of Android malware. Raju et al. [6] explain that the rapid growth of mobile devices has made the Android operating system one of the primary targets for malware attacks. Android malware continues to evolve with increasingly complex attack techniques, necessitating more adaptive and automated analytical approaches. In this context, NLP is used to assist in the analysis of cyber threat reports related to Android malware to generate intelligence datasets that can be utilized in the development of malware detection models based on machine learning and deep learning.The development of threat intelligence-based datasets is a key focus of this research. According to studies by Rahman et al. [7] and Xu et al. [8], the integration of NLP, artificial intelligence, and cybersecurity offers significant opportunities for building smarter and more automated cybersecurity systems. Datasets generated through automated extraction from cybersecurity reports are expected to provide richer intelligence context compared to conventional malware datasets. The extracted data is then used in classification, prediction, and threat pattern analysis processes using various artificial intelligence and deep learning approaches.In its implementation, the research also utilizes various machine learning and deep learning approaches to support natural language analysis. Ainslie et al. [9] explain that techniques such as text classification, topic modeling, sentiment analysis, named entity recognition (NER), and relation extraction play a crucial role in understanding the structure and context of text data more deeply. Furthermore, the development of transformer-based language models and Large Language Models opens new opportunities for the development of artificial intelligence-based cyber threat analysis systems capable of performing analysis automatically and in real-time.The application of NLP is also utilized in various other fields such as business intelligence, social media analysis, data analytics, and decision support systems. According to Ismail’s research [4], NLP can be used to analyze customer opinions, understand market trends, and identify user behavior patterns from available digital data. In social media analysis, NLP is utilized to perform sentiment analysis, topic detection, and mapping of public opinion regarding specific issues. This demonstrates that NLP plays a crucial role in supporting data-driven digital transformation across various industrial sectors and modern organizations.Research development in the field of NLP is also directed toward supporting the processing of the Indonesian language, which still faces various challenges compared to languages with abundant resources, such as English. These challenges include limited datasets, variations in language structure, the use of informal language, and a scarcity of available linguistic resources. Therefore, research in the field of NLP is expected to contribute to the development of natural language processing technologies that are more adaptable to the characteristics of the Indonesian language, particularly in the domains of cybersecurity and intelligence analysis.Looking ahead, the direction of expertise development is focused on the integration of NLP, artificial intelligence, cyber intelligence, and Large Language Models to build smarter, automated, and context-aware cyber threat analysis systems. According to Ahi and Valizadeh [10], the use of LLMs in the field of cybersecurity has great potential to enhance a system’s ability to understand threat contexts, reason about cyber threat data, and generate more accurate and relevant intelligence information. With these technological advancements, NLP is expected to become a key foundation in supporting digital transformation, strengthening cybersecurity, and developing data-driven intelligent systems in the future.References [1] H. Xue and W. Liu, “Bibliometric Analysis of Natural Language Processing Technology in Education: Hot Topics, Frontier Evolution, and Future Prospects,” SAGE Open, vol. 15, no. 1, 2025, doi: 10.1177/21582440251319891. [2] M. A. Ferrag et al., “Revolutionizing Cyber Threat Detection with Large Language Models: A Privacy-Preserving BERT-Based Lightweight Model for IoT/IIoT Devices,” IEEE Access, vol. 12, pp. 23733–23750, 2024, doi: 10.1109/ACCESS.2024.3363469. [3] M. Arazzi et al., “NLP-based techniques for Cyber Threat Intelligence,” Computer Science Review, vol. 58, p. 100765, 2025, doi: 10.1016/j.cosrev.2025.100765. [4] W. S. Ismail, “Threat Detection and Response Using AI and NLP in Cybersecurity,” Journal of Internet Services and Information Security, vol. 14, no. 1, pp. 195–205, 2024, doi: 10.58346/JISIS.2024.I1.013. [5] M. Albarrak, K. Salonitis, and S. Jagtap, “Natural Language Processing (NLP)-Based Frameworks for Cyber Threat Intelligence and Early Prediction of Cyberattacks in Industry 4.0: A Systematic Literature Review,” Applied Sciences, vol. 16, no. 2, p. 619, 2026, doi: 10.3390/app16020619. [6] A. D. Raju, I. Y. Abualhaol, R. S. Giagone, Y. Zhou, and S. Huang, “A Survey on Cross-Architectural IoT Malware Threat Hunting,” IEEE Access, vol. 9, pp. 91686–91708, 2021, doi: 10.1109/ACCESS.2021.3091427. [7] M. A. Rahman et al., “A Survey of Large Language Models (LLMs) for Cybersecurity: Opportunities and Directions,” in Proc. 2025 IEEE Int. Big Data Conference (BigData), 2025, pp. 4333–4342, doi: 10.1109/BigData66926.2025.11402639. [8] H. Xu et al., “Large Language Models for Cybersecurity: A Systematic Literature Review,” ACM Transactions on Software Engineering and Methodology, 2025, doi: 10.1145/3769676. [9] S. Ainslie, D. Thompson, S. Maynard, and A. Ahmad, “Cyber-threat intelligence for security decision-making: A review and research agenda for practice,” Computers and Security, vol. 132, 2023, doi: 10.1016/j.cose.2023.103352. [10] K. Ahi and S. Valizadeh, “Large Language Models (LLMs) and Generative AI in Cybersecurity and Privacy: A Survey of Dual-Use Risks, AI-Generated Malware, Explainability, and Defensive Strategies,” in 2025 Silicon Valley Cybersecurity Conference (SVCC), 2025, pp. 1–8, doi: 10.1109/SVCC65277.2025.11133642.

Volume 1, Issue 2 (2026)

Articles in This Issue