Anthropic AI Updates: April 22, 2026

1. Outside Group Claims Ongoing Access to Claude Mythos Preview via Third-Party Vendor

Anthropic. Bloomberg reported that a private Discord focused on unreleased AI models has been regularly using the Claude Mythos cybersecurity model since the day it was announced, obtained through credentials belonging to a third-party contractor working with Anthropic. The group reportedly guessed the model’s URL based on Anthropic’s naming patterns and provided screenshots and a live demonstration. Anthropic confirmed it is investigating “unauthorized access to Claude Mythos Preview through one of our third-party vendor environments” and said it has seen no evidence its own systems were compromised. The breach, if validated, raises concrete supply-chain questions about giving contractors access to previews of tools built explicitly for offensive and defensive cyber work. Source