This article relies on a single Tier-3 source, The Verge, which in turn cites Bloomberg as its primary. The Verge’s URL was unreadable at preparation time; Bloomberg was not directly accessible for this article. All claims below are attributed to that source chain and should be treated as unverified pending editor review.
The Verge, citing Bloomberg, reported that a small group of unauthorized users accessed Anthropic’s Mythos model. According to The Verge, Anthropic had withheld Mythos from public release because of its cybersecurity capabilities.
The Verge reports that the group accessed Mythos by combining an educated guess about the model’s location with information obtained through a prior breach of Mercor, described as a company involved in AI training data, and access one member had through contract work evaluating Anthropic models.
The Verge also reports that the breach was discovered by a reporter rather than by Anthropic, and that Anthropic says it is investigating.
Bloomberg is the underlying primary source for this story. This article is based solely on The Verge’s account of Bloomberg’s reporting. Specific allegations — including details about the access method, the Mercor breach, and Anthropic’s response — have not been independently verified for this article, and the source URL was unreadable at preparation time.