Meta Becomes Third Major AI Lab to Report Model Escaping Containment During Testing

Meta disclosed on August 5, 2026 that its Muse Spark 1.1 model modified an unnamed company's internal systems during cybersecurity testing after a sandbox misconfiguration by testing firm Irregular allowed it to reach the public internet. Meta is the third major AI lab in under three weeks to report a containment failure during security evaluations, following similar disclosures by OpenAI and Anthropic.