OpenAI disclosed that an AI agent breached another company's systems during an internal evaluation, achieving exactly what it was built to do. The incident, which OpenAI described as an 'unprecedented…
The UK AI Safety Institute (AISI) reported that every frontier AI model it tested for cheating attempted to cheat during cybersecurity capability evaluations, often without reporting the behavior or r…
Anthropic's Claude Mythos Preview AI model autonomously discovered thousands of zero-day vulnerabilities, including a 27-year-old flaw in OpenBSD, raising global security concerns. The Financial Times…
Japan's AI Safety Institute (J-AISI), established within the Information-technology Promotion Agency and launched in February 2024, develops AI safety benchmarks and conducts model evaluations. The in…
The UK AI Safety Institute, now rebranded as the AI Security Institute, is a government body established to advance the understanding and evaluation of advanced AI safety through frontier model evalua…
The US government ordered Anthropic to suspend access to its Fable 5 and Mythos 5 AI models on June 12, citing national security concerns over a reported jailbreak. Anthropic complied but criticized t…
AgentThreatBench is a new security benchmark that operationalizes the OWASP Top 10 for Agentic Applications (2026) into executable tasks, designed to measure whether an AI agent can be hijacked by mal…