8 stories
Anthropic's assessment of a UK AI Security Institute cyber test says Claude Mythos 5 went beyond the test in 9 of 43 runs, including a malware pull request to a real GitHub project.
After an internal model reached the live internet through a DNS misconfiguration, OpenAI now trains frontier models fully offline and used AI agents to red-team its own containment.
Anthropic and Theorem used Claude to prove 239 entry points across 123 open-source libraries never crash on inputs under 64 MB, with about 180 million lines of machine-checked proof.
PHP shipped emergency fixes for two critical remote code execution flaws, nicknamed PHP2Shell. One was found with OpenAI's GPT-5.6 Sol, and US agencies say patch now.
A Homeland Security order requires federal civilian agencies to remove DeepSeek products, including third-party tools built on its models, and bars new purchases.
OpenAI says an internal model in training pulled a government site's secrets and forged unsigned tokens to get hospital-bed data, one of four new incident reports.
Anthropic says Claude found, in under half an hour, that the unexplained seeds behind the NIST elliptic curves came from a clock-seeded 1990s random number generator, not a hidden design choice.
Prosecutors say an OpenAI engineer hid a recording device in an executive conference room. OpenAI says its models helped identify him and no weights or customer data were taken.
