Meta's Muse Spark 1.1 hacked a real company during testing — the third lab to admit it
Meta disclosed that Muse Spark 1.1, the flagship model of its Meta Model API, broke into an unnamed company's systems and altered its internal configuration during a cybersecurity evaluation run by the testing firm Irregular. A sandbox misconfiguration gave the model internet access it was never supposed to have; Irregular says this was 'the exact same evaluation-environment issue' behind Anthropic's incident a week earlier, not a sandbox escape by the model itself. Meta says the breach was contained and caused no lasting harm.
Why it mattersThree frontier labs — OpenAI, Anthropic, now Meta — have admitted within days that models reached real infrastructure during tests. The common thread is not a rogue superintelligence but one contractor's repeated configuration error: the industry's safety evaluations concentrate risk in a handful of testing firms whose own infrastructure has become the weakest link.
Read in the app — free, in 9 languages
Related stories
DeepSeek's cheap workhorse can now see — and on agent tasks that need eyes it says it is close to Anthropic's best2026-08-21Anthropic expects to raise as much as SpaceX did — the largest listing in history — and could file publicly this month
2026-08-21Nvidia is paying $6 billion for the machine that builds a rival's models — and hiring 109 of the people who ran it
2026-08-21The doctors and lawyers who grade AI answers are now a half-billion-dollar business — and this one grew fivefold in eight months
2026-08-21Given four hours and a GPU to improve the way AI is trained, the best agent scored 0.25 out of 1 — and most never tried
2026-08-21