23/04/2026
BREAKING π¨ π± Anthropic is looking into reports that a small group of unauthorized users gained access to Claude Mythos Previewβa highly advanced AI model that was never intended for public release because of its powerful cybersecurity capabilities.
The access does not appear to stem from a direct breach of Anthropicβs core systems. Instead, itβs believed to have come through a third-party vendor environment.
Key Details
How it happened:
A small group in a private online forum reportedly accessed the model using compromised contractor credentials, along with educated guesses about where the model was hosted. Some of the clues may have come from a separate data breach at Mercor.
What they did:
According to Bloomberg, the group appears to be AI hobbyists who used the model for low-risk, non-malicious tasksβlike building simple websitesβto stay under the radar.
Anthropicβs response:
An Anthropic spokesperson says the company is actively investigating the third-party environment and, so far, has found no evidence that its internal systems were compromised.