Meta said a misconfiguration by the independent testing company Irregular inadvertently allowed one \u2060of its models internet access during an evaluation, adding that it was investigating the incident.<\/p>\n
The model \u201cexploited \u200ba security vulnerability in a third-party service, \u200cin a manner similar \u200cto previously reported instances with other companies\u201d, Meta said in a statement.<\/p>\n
Earlier in the day, \u200cThe Information, citing sources, reported that Meta\u2019s Muse Spark 1.1 model, which it has touted as its most capable model for real-world coding and agentic tasks, breached an unidentified company and altered its internal systems.<\/p>\n
A spokesperson for Irregular told Reuters the incident was the \u201cexact same evaluation-environment issue that was already disclosed by Anthropic last week\u201d and that it did \u200cnot involve a \u201csandbox escape or a sophisticated cyber action\u201d.<\/p>\n
\u201cThere are no current open issues. Irregular is developing a white paper to share best \u200bpractices for containment and securely running cyber evaluations,\u201d Irregular said.<\/p>\n
The incidents revealed by Meta and Anthropic were due to mistakes that inadvertently gave their models access to the open internet. That contrasts with OpenAI, whose AI agent independently exploited a novel vulnerability to reach the internet during cyber testing.<\/p>\n
![\"\"](\"https:\/\/www.newsbeep.com\/ie\/wp-content\/uploads\/2026\/06\/1782832089_715_500.jpg\")
<\/p>\n