Meta confirms: AI model hacked another company
The incident occurred during cybersecurity testing due to an error by independent firm Irregular, with similar issues reported by OpenAI and Anthropic.
The incident occurred during cybersecurity testing due to an error by independent firm Irregular, with similar issues reported by OpenAI and Anthropic.
On Wednesday, August 6, 2026, Meta reported that during security testing, one of its AI models exploited a vulnerability and gained unauthorized access to another company's systems. The incident has sparked a new wave of concern over the growing security risks posed by advanced AI systems.
Meta claims the incident was caused by a configuration error in the test environment, for which external firm Irregular was responsible. Due to this error, the model was given internet access during testing, and it then exploited a vulnerability in a third-party system.
The Information portal, citing sources, reported that the model in question was Muse Spark 1.1, which Meta claims is its most advanced model for coding and executing complex tasks. The model breached the defenses of an unknown company and altered part of its internal computer network.
On the same day, Meta also announced on Twitter the release of the new Muse Spark 1.2 model, which powers Muse Code, a terminal coding agent for complex software tasks. "Releasing Muse Code in beta today. It's a terminal coding agent that takes on complete software engineering tasks across large repos: planning changes, writing code, validating the results. Powered by Muse Spark 1.2, a coding-focused model update", wrote Mark Zuckerberg, Meta's founder and CEO. It has not been confirmed whether the incident is linked to testing of the new model version.
An Irregular spokesperson told Reuters that this was the same testing oversight reported by Anthropic last week. They emphasized that this was not an escape from a secure environment or a sophisticated cyberattack, but rather the result of a misconfiguration.
A Meta spokesperson confirmed to the BBC that the company is investigating the hack caused by a "misconfiguration" and described it as similar to previously reported incidents at other companies. Meta also announced it would release more information "once we have all the facts".
Similar security issues have been reported at other companies developing artificial intelligence. In the past two weeks, OpenAI and Anthropic have also reported incidents where their models hacked systems of other organizations during testing. OpenAI's agents attacked several publicly available services, including the AI tool platform Hugging Face, while Anthropic's Claude model carried out similar attacks after a misconfiguration gave it internet access.
Some commentators have questioned the timing of these incident disclosures, given that tech companies are vying for dominance in AI development. OpenAI and Anthropic are preparing for stock market listings, with each expected to be valued at around one trillion dollars (approximately 740 billion pounds).
Following its own security incident, OpenAI announced the release of a technical report, while a group of Republican state attorneys general asked the company to preserve all documentation related to the case. Earlier this week, the White House convened a meeting with leading AI companies to discuss a new protocol for security testing of advanced AI systems.
This week, the UK's AI Safety Institute (AISI) announced that its testing showed some models attempted cyberattacks by creating fake human profiles. In the most serious case, Anthropic's Mythos AI model tried to gain access to a service by sending private messages from fake accounts impersonating real people. Anthropic stated that AISI's tests were not "representative of any of our production models", while OpenAI said the evaluations did not reflect typical usage.
