Unauthorized Access Hits Anthropic's Restricted Mythos AI
A limited number of people without permission entered Anthropic's latest AI system, Claude Mythos. Bloomberg shared this information. The company views this model as capable of supporting serious cyber threats. For that reason, Anthropic restricts access to trusted collaborators such as Apple, Amazon, and Cisco. They provide it through the Project Glasswing initiative.
How the Breach Occurred
Members of a private Discord server managed to log in on the announcement date. One participant works as a contractor for Anthropic. The group combined that person's login details with data exposed in a leak from Mercor, an AI startup. Bloomberg notes the intruders stuck to safe activities. They created basic websites for experiments. No harmful cyber operations took place.
Anthropic started in 2021. Former OpenAI staff founded it to prioritize AI safety. The Claude family of models powers many applications today. Mythos stands out for its advanced abilities. The firm limits it to prevent misuse in attacks on networks or systems.
Additional Model Access
The same individuals reached other upcoming Anthropic AI systems. Details on those remain unclear. The company confirmed an ongoing probe into the matter. Current findings show the entry stayed within the contractor's setup. Nothing points to intrusion into Anthropic's core infrastructure.
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
Project Glasswing connects Anthropic with major tech firms. Apple integrates Claude in devices. Amazon uses it for cloud services. Cisco applies it in networking tools. These partnerships highlight trust in Anthropic's controls. Yet this event raises questions about contractor security.
Mercor focuses on matching AI talent with jobs. Its leak made credentials public. Such incidents remind the industry of leak risks. Anthropic now reviews access protocols. Contractors handle sensitive tasks often. Strong verification protects models like Mythos.
Company Background and Implications
Anthropic builds AI with safety first. Claude models compete with leaders like GPT. Mythos pushes boundaries in power. Restrictions aim to block dangers. The Discord group found a weak point. Their actions stayed innocent so far.
No evidence shows wider damage. Anthropic acts quickly. The firm shares updates as they come. Partners stay informed. This case underscores contractor vetting needs. AI firms tighten rules amid growth.
Bloomberg first reported the details. The story broke on April 22, 2026. Watch for Anthropic statements.
Related on Neura Market