Хронология самых интересных ИИ релизов в 2024 году🔥

От Gemma до Llama 3.1 405B, от Sonnet 3.5 до o3 !

https://huggingface.co/spaces/reach-vb/2024-ai-timeline

@data_analysis_ml

🔐 AI/LLM Security Update
🎯 Query: LLM security vulnerability
━━━━━━━━━━━━━━━━━━━━━

📌 Targeting the Core: A Simple and Effective Method to Attack RAG ...
└─ 🔗 Source
└─ 📝 Dec 5, 2024 ... ... methodology, revealing systemic vulnerabilities even in robust architectures. These studies underscore the urgency of developing attack-resilient LLM frameworks ...

📌 The Task Shield: Enforcing Task Alignment to Defend Against ...
└─ 🔗 Source
└─ 📝 Dec 21, 2024 ... The Task Shield: Enforcing Task Alignment to Defend Against Indirect Prompt Injection in LLM Agents ... vulnerabilities to prompt injection attacks while ...

📌 Trust & Safety of LLMs and LLMs in Trust & Safety
└─ 🔗 Source
└─ 📝 Dec 4, 2024 ... ... prompt injection exploits the LLM's inability to distinguish between legitimate instructions and malicious ones. This vulnerability is particularly ...

📌 The Task Shield: Enforcing Task Alignment to Defend Against ...
└─ 🔗 Source
└─ 📝 Dec 21, 2024 ... vulnerable to prompt injection attacks, making it an ideal candidate to ... Prompt injection attack against llm-integrated appli- cations. arXiv ...

📌 arXiv:2405.03644v2 [cs.CR] 4 Dec 2024
└─ 🔗 Source
└─ 📝 Dec 4, 2024 ... ACM Transactions on Software Engineering and Methodology, July. 2024. ... Strengthening llm trust boundaries: A survey of prompt injection attacks surender suresh ...

📌 Pirates of the RAG: Adaptively Attacking LLMs to Leak Knowledge ...
└─ 🔗 Source
└─ 📝 8 days ago ... (2024) , which we refer to as Prompt-Injection for Data Extraction (PIDE), follows a related methodology but instead draws its textual inputs from the WikiQA ...

📌 Comprehensive Assessment of Jailbreak Attacks Against LLMs
└─ 🔗 Source
└─ 📝 Dec 16, 2024 ... Adversaries exploit these capabilities to launch a variety of attacks, notably prompt injection ... Content Analysis: An Introduction to Its Methodology.

📌 arXiv:2411.00459v2 [cs.CR] 23 Dec 2024
└─ 🔗 Source
└─ 📝 Dec 23, 2024 ... 4 Methodology. 4.1 Problem Formulation. Given an input instruction I and clean ... Optimization-based prompt injection attack to llm- as-a-judge. arXiv ...

📌 Chain-of-Scrutiny: Detecting Backdoor Attacks for Large Language ...
└─ 🔗 Source
└─ 📝 Dec 21, 2024 ... 3 Methodology. Report issue for preceding element. 3.1 Threat Model and ... Effectiveness of Prompt-injection based Backdoor Attacks. Following the ...

Правила детекирования sigma,splunk с маппингом на техники mitre
https://attackrulemap.netlify.app/

#MLSecOps
"CySecBench: Generative AI-based CyberSecurity-focused Prompt Dataset for Benchmarking Large Language Models", 2025.

Taxonomy of attacks in Red-Teaming Large Language Models (LLMs). Attacks range from prompt-based
attacks, which only require access to the application prompt, to training attacks, which require access to instruction tuning or fine-tuning data and the training process.

Взято из статьи "Operationalizing a Threat Model for Red-Teaming Large Language Models (LLMs)"