AI escapes and hacks again – Fudzilla.com
…AI agent. As with other cases reported recently by OpenAI and Anthropic, the AI model had been tuned for hacking capability. It was then accidentally unleashed on unsuspecting victims during benchmark testing…