OpenAI Opens GPT-5.5 Bio Bug Bounty to Test Frontier AI Safety Controls
OpenAI is inviting vetted red-teamers, security researchers, and biosecurity specialists to test whether GPT-5.5 can be universally jailbroken on five biology safety questions. The private bounty offers $25,000 for the first true universal jailbreak.
Anthropic Unveils Project Glasswing to Put Frontier AI on Cyber Defense Duty
Anthropic has launched Project Glasswing, a new initiative built around Claude Mythos Preview to help secure critical software before advanced AI systems make cyberattacks easier to scale. The company is framing it as a defense-first response to rapidly improving AI vulnerability research.
Claude Mythos Leaked: Anthropic's Most Powerful Model Yet Poses 'Unprecedented Cybersecurity Risk'
A CMS misconfiguration exposed nearly 3,000 internal Anthropic assets, including a draft blog post describing Claude Mythos — a new model tier above Opus that the company itself warns is 'far ahead of any other AI model in cyber capabilities.' Anthropic has confirmed the model exists.
Stanford Study: AI Chatbots Are Making People More Self-Centered — and Tech Companies Have Every Reason to Keep It That Way
A peer-reviewed Stanford paper published in Science finds that AI sycophancy doesn't just flatter users — it makes them less likely to apologize, more morally rigid, and increasingly dependent on machine validation over human feedback. The researchers say it's a safety issue that needs regulation.
OpenAI Launches Safety Bug Bounty Program to Reward Researchers Who Find AI Abuse Risks
OpenAI is opening a public Safety Bug Bounty program targeting AI-specific misuse scenarios — from agentic prompt injection to platform integrity bypasses — that fall outside traditional security vulnerability scopes.