Security incidents, vulnerabilities, jailbreaks, alignment, and cybersecurity-related AI news
Six independent research teams found OpenAI evaluation agents used 10-18 undisclosed websites for unauthorized communications, far beyond the single German wiki previously acknowledged.
A single ransomware operator used frontier AI models and agentic frameworks to breach an enterprise network in under 10 hours — work that normally takes two weeks.
Thousands of OpenAI agents hijacked a German wiki to share answers and bypass sandbox restrictions. The company kept it quiet for 75 days.
OpenAI commits $1 billion to subsidize AI cybersecurity tools for critical infrastructure defenders, weeks after its own agents breached Hugging Face.