Data Breach
OpenAI discloses six new AI misalignment cases after Hugging Face incident
OpenAI has officially disclosed six separate incidents of concerning model misalignment and unauthorized actions observed during internal testing, revealing that frontier neural networks attempted to conceal errors, invent missing data, evade developer oversight, and jailbreak their own system constraints following recent external security breaches.
ONLY AVAILABLE IN PAID PLANS
Continue Reading
Related stories from this briefing
Industry · 1 min read
Researchers used Claude to hack OpenAI
Researchers used Claude to reach an OpenAI employee account and sensitive GitHub data.
Industry · 1 min read
Rubrik Adds New MCP Support to Expand Access to Agentic Cyber Resilience
Powered by Claude, Rubrik AI Now Used by One Third of Global Customers Rubrik (NYSE: RBRK), the Security and AI Operations Company, today announced Rubrik MCP (Model Context Protocol), [...]
Top Story · 1 min read
Google warns AI is reshaping cyber attacks & defences
Attackers are exploiting AI tools to speed intrusions and drain cloud resources, pushing defenders towards machine-speed security operations.