- Sirdaryo viloyati Guliston shahar
- Dushanba-Shanba: 9.00 - 18.00
Xaitbayev Azizbek Pirnazarovich
10.5281/zenodo.22938593
Annotatsiya
Generativ sun’iy intellekt xizmatlari matn, rasm, kod va boshqa kontentni yaratish imkonini beradi. Biroq bu xizmatlardan foydalanganda maxfiy ma’lumotlar noto‘g‘ri tarzda modelga kiritilishi yoki model javoblarida bilvosita “oqib chiqishi” xavfi mavjud. Shuning uchun GenAI tizimlarida maxfiy ma’lumotlarni aniqlash va himoyalash xavfsizlik strategiyasining ajralmas qismi hisoblanadi. Ushbu annotatsiya GenAI xizmatlarida maxfiy ma’lumotlarni aniqlashning amaliy va samarali usullarini ko‘rib chiqadi hamda xavfsizlikka alohida urg‘u beradi. Maxfiy ma’lumotlarni aniqlash jarayoni odatda uch bosqichga ajraladi: (1) foydalanuvchi kiritishi (prompt) va fayllarni tekshirish, (2) model ichida ishlov berish vaqtida xavflarni kamaytirish, (3) model javoblarini nazorat qilish va potensial sizib chiqishni bloklash. Bunday yondashuv “ma’lumot kiritishdan boshlab himoya” tamoyiliga asoslanadi va xavfsizlikni tizimli holatga keltiradi. 1. Statik va semantik tahlil asosida aniqlash (Rule-based + Pattern Matching) Eng sodda usullar — maxfiy ma’lumotlarga xos naqshlar (pattern) va formatlarni aniqlashdir. Masalan, elektron pochta manzillari, telefon raqamlari, pasport/ID formatlari, karta raqamlari, soliq kodlari, parol yoki API kalitlarining tipik ko‘rinishlari regex (muntazam ifodalar) orqali topilishi mumkin. Bunday usullar tez ishlaydi, interpretatsiya oson va “yuqori aniqlik” talab qilinadigan joylarda foydalidir. Shu bilan birga, faqat regexga tayanish chetlab o‘tish (obfuscation) holatlarida yetarli bo‘lmasligi mumkin (masalan, raqamlar orasiga bo‘shliq qo‘shish, qisman yashirish). Shu sababli pattern matching semantik tahlil bilan birga qo‘llanishi kerak.
Kalit so'zlar:
Foydalanilgan adabiyotlar
1. Greshake, K., Abdelnabi, S., Mishra, S., Endres, C., Holz, T., & Fritz, M. (2023).
Not what you’ve signed up for: Compromising Real-World LLM-Integrated
Applications with Indirect Prompt Injection. Proceedings of the 16th ACM
Workshop on Artificial Intelligence and Security (AISEC ’23), 79–90.
2. Yao, Y., Duan, J., Xu, K., Cai, Y., Sun, Z., & Zhang, Y. (2024). A Survey on Large
Language Model (LLM) Security and Privacy: Vulnerabilities, Attacks, and
Defenses. IEEE Transactions on Knowledge and Data Engineering (TKDE), 1–20.
3. Inan, H., Upasani, K., Chi, J., Rungta, R., Iyer, K., Mao, Y., Tontchev, M., Hu, Q.,
Fuller, B., & Schelter, S. (2023). Llama Guard: LLM-based Input-Output Safeguard
for Human-AI Conversations. arXiv preprint arXiv:2312.06674.
4. Honovich, O., Scialom, T., Levy, O., & Schick, T. (2022). Unnatural Instructions:
Tuning Language Models with (Almost) No Human Labor. ACL 2023 / arXiv
preprint arXiv:2212.09689. (Embedding hamda Named Entity Recognition (NER)
usullari bilan text screening yondashuvlari bo’yicha)
5. NIST (National Institute of Standards and Technology). (2023). Artificial
Intelligence Risk Management Framework (AI RMF 1.0). NIST Trustworthy and
Responsible AI, NIST SP 1270, Washington, D.C.
6. Mink, J., Yu, Z., & Carlini, N. (2024). Data Loss Prevention for Large Language
Models: Technical Challenges and Solutions. IEEE Security & Privacy, 22(2), 45
54.
7. Reimers, N., & Gurevych, I. (2019). Sentence-BERT: Sentence Embeddings using
Siamese BERT-Networks. Proceedings of the 2019 Conference on Empirical
Methods in Natural Language Processing (EMNLP), 3982–3992.