#MLSecOps
#Offensive_security
"Multi-Faceted Attack: Exposing Cross-Model Vulnerabilities in Defense-Equipped Vision-Language Models", Nov. 2025.

// Multi-Faceted Attack (MFA) - framework that systematically uncovers general safety vulnerabilities in leading defense-equipped VLMs, including GPT-4o, Gemini-Pro, and LlaMA 4, etc. Central to MFA is the Attention-Transfer Attack, which conceals harmful instructions inside a meta task with competing objectives. We offer a theoretical perspective grounded in reward-hacking to explain why such an attack can succeed

#AIOps
#CogSec
#MLSecOps
"Cognitive Control Architecture (CCA): A Lifecycle Supervision Framework for Robustly Aligned AI Agents", Dec.2025.

// Method is predicated on a core insight: no matter how subtle an IPI attack, its pursuit of a malicious objective will ultimately manifest as a detectable deviation in the action trajectory, distinct from the expected legitimate plan

See also:
]-> Dynamic Environment to Evaluate Prompt Injection Attacks and Defenses for LLM Agents
]-> https://agentdojo.spylab.ai

#MLSecOps
#Threat_Modelling
"Cisco Integrated AI Security and Safety Framework Report", Dec. 2025.
]-> Cisco AI security and safety taxonomy

// This paper presents Cisco’s Integrated AI Security and Safety Framework, a unified, lifecycleaware taxonomy and operationalization framework that can be used to classify, integrate, and operationalize the full range of AI risks. It integrates AI security and AI safety across modalities, agents, pipelines, and the broader ecosystem

#MLSecOps
"Safe2Harm: Semantic Isomorphism Attacks for Jailbreaking Large Language Models", Dec. 2025.

// Most existing jailbreak methods revolve around Prompt Engineering or adversarial optimization, yet we identify a previously overlooked phenomenon: many harmful scenarios are highly consistent with legitimate ones in terms of underlying principles. Based on this finding, this paper proposes the Safe2Harm Semantic Isomorphism Attack method

#AIOps
#MLSecOps
#RAG_Security
"MemoryGraft: Persistent Compromise of LLM Agents via Poisoned Experience Retrieval", 2025.
]-> https://github.com/Jacobhhy/Agent-Memory-Poisoning

// MemoryGraft - novel indirect injection attack that compromises agent behavior not through immediate jailbreaks, but by implanting malicious successful experiences into the agent's long-term memory. Unlike traditional prompt injections that are transient, or standard RAG poisoning that targets factual knowledge, MemoryGraft exploits the agent's semantic imitation heuristic which is the tendency to replicate patterns from retrieved successful tasks

#MLSecOps
#Red_Team_Tactics
"From Rookie to Expert: Manipulating LLMs for Automated Vulnerability Exploitation in Enterprise Software", Dec. 2025.
]-> All data, source code, and instructions

// We show in this work how publicly available LLMs can be socially engineered to transform novices into capable attackers, challenging the foundational principle that exploitation requires technical expertise. To that end, we propose RSA (Role-assignment, Scenario-pretexting, and Action-solicitation), a pretexting strategy that manipulates LLMs into generating functional exploits despite their safety mechanisms

#MLSecOps
#Whitepaper
"Implementing Secure AI Framework Controls in Google Cloud", Dec. 2025.

// Google’s Secure AI Framework is a framework for securing AI systems throughout their lifecycles. SAIF is designed for practitioners – the security professionals, developers, and data scientists on the front lines – to ensure AI models and applications are secure by design

DevSecOps Assessment Framework для ML от "Инфосистемы Джет"

🔍 Что это и зачем это нужно?

Недавно команда "Инфосистемы Джет" выпустила DevSecOps Assessment Framework для ML. Это подробная модель зрелости процессов безопасной разработки систем искусственного интеллекта.

В ней структурированный чек-лист/дорожная карта, которая помогает понять, на каком уровне находится ваша организация в части MLSecOps: от базового контроля зависимостей до продвинутой защиты от специфических угроз ИИ.

🛡 Основные положения фреймворка

• Контроль зависимостей, артефактов и SBOM/ML-BOM
• Защита сред разработки, секретов, CI/CD и SCM
• Анализ и очистка обучающих/валидационных данных (отравление, PII, jailbreak в RAG)
• Динамическое тестирование LLM на jailbreak, prompt injection и другие атаки
• Мониторинг, аудит и реагирование в продакшене
• Обучение команд требованиями ИБ

В документе в качестве примеров полезных инструментов для поиска уязвимостей в LLM упоминаются HiveTrace Red и Llamator — приятный комплимент инструментам сообщества 🤝

Если тема AI Security вам близка, рекомендуем изучить и использовать как чек-лист или дорожную карту для улучшений.

Ссылка на релиз:
https://github.com/Jet-Security-Team/DevSecOps-Assessment-Framework/releases/tag/2025.12.26

#MLSecOps #AISecurity #DevSecOps #LLMsecurity

Разбор Никиты Беляевского, разработчика HiveTrace Red (https://hivetrace.ru/red)

#Research
#MLSecOps
"A Safety Report on GPT-5.2, Gemini 3 Pro, Qwen3-VL, Doubao 1.8, Grok 4.1 Fast, Nano Banana Pro, and Seedream 4.5", 2026.
]-> Repo

// This report presents a comprehensive safety evaluation of the latest foundation models released in 2026, including GPT-5.2, Gemini 3 Pro, and others. We analyze safety alignment across text, vision-language, and text-to-image modalities, highlighting vulnerabilities in current safeguards against adversarial attacks and regulation compliance

#MLSecOps
"FNF: Functional Network Fingerprint for Large Language Models", Feb. 2026.
]-> Repo

// In this work, we propose the Functional Network Fingerprint (FNF), a training-free, sample-efficient method for detecting whether a suspect LLM is derived from a victim model, based on the consistency between their functional network activity