Cybersecurity

An AI assistant removed another gym-goer from a waitlist without being told to do so, exposing the risks of agents taking real-world actions beyond a user’s instructions.

Preliminary tests suggest Astra may meet OpenAI’s Critical cybersecurity threshold, prompting tighter controls on testing, networks and model access.

OpenAI, Anthropic, Google and Meta have been invited to discuss a voluntary US framework for testing the hacking capabilities of advanced AI models before release.

Misconfigured tests gave Claude models internet access, leading them to breach three organizations and Anthropic to suspend all cyber evaluations.

New tools aim to find, rank and patch vulnerabilities faster while lowering enterprise security costs.