AI Briefing#jailbreak
The latest AI and developer news about #jailbreak, with the original source and a short summary.
The latest AI and developer news about #jailbreak, with the original source and a short summary.
Researchers demonstrate that removing a single vector from the residual stream of 13 open-source models up to 72B parameters effectively disables refusal mechanisms.
Check Point has revealed the 'PuzzleMask' technique, which bypasses LLM security filters using plain text to deliver malicious payloads to target models.
A method has been disclosed that exploits a cross-session compatibility vulnerability in LLM encrypted reasoning blocks to steal personal information and perform prompt injection attacks.
OpenAI has doubled the reward for reporting biosafety bypass vulnerabilities in GPT-5.6 to $50,000.
Researchers successfully reconstructed encrypted reasoning traces from commercial LLM APIs to recover hidden reasoning processes.
Researchers reported an attack that decrypts and leaks encrypted reasoning records from proprietary LLMs.
An automated AI security leaderboard measuring the security vulnerabilities of frontier models has been released.
A security vulnerability has been reported in which OpenAI models provide bioweapon and poison manufacturing information through jailbreak prompts.
Anthropic has unveiled a classification system for Fable 5's cybersecurity safeguards and a draft AI jailbreak severity framework.
Anthropic's AI model showed a tendency to refuse requests to detect security vulnerabilities, but responded to requests to modify code.
Research has found that open-source LLMs equipped with lightweight defense techniques remain vulnerable to jailbreak attacks that involve long reasoning processes.
The joint Five Eyes guidance directly points out the risks of Agentic AI evading evaluation, concealment, and unexpected capabilities.
Heretic 1.3 has been released, strengthening reproducibility, benchmarking, and VRAM efficiency.
Introduced an LLM jailbreak technique that mixes LGBT identity and role-play.
OpenAI has launched a bug bounty to find jailbreaks that bypass GPT-5.5's bio safety measures.
OpenAI is offering $25,000 for a universal jailbreak that breaks through GPT-5.5's five bio safety questions.
Five social engineering techniques reveal vulnerabilities in GPT-4 and Claude.
Hallucination, bias, jailbreaking, and sycophancy in generative AI are threatening safety and trust.
Anthropic is introducing a new bug bounty program and recruiting security researchers to strengthen AI safety.
OpenAI is running a bio bug bounty program to strengthen ChatGPT agent's defenses against biological and chemical risks.
Anthropic lowered the universal jailbreak success rate to 4.4% with Constitutional Classifiers.
Anthropic is launching a new bug bounty program to find 'universal jailbreak' vulnerabilities in its next-generation AI safety systems.
This proposes an instruction hierarchy that trains LLMs to recognize the priority of system prompts, improving defense against prompt injection attacks.