Unauthorized Access to Anthropic’s AI Model
Unauthorized users gained access to Anthropic’s restricted Mythos AI model on launch day via a third-party contractor’s environment.
April 22, 2026 – 7:27 am
A small group communicating via a private Discord channel accessed Claude Mythos Preview by guessing the model’s URL on the same day Anthropic announced Project Glasswing. Anthropic states it is investigating and has found no evidence of impact to its core systems. The breach highlights the risks of restricting access to advanced AI capabilities through vendor environments rather than technical controls.
A small group of unauthorized users gained access to Anthropic’s closely restricted cybersecurity AI model, Claude Mythos Preview, on the same day the company publicly announced it, apparently by guessing the model’s URL based on knowledge of Anthropic’s URL format for other models, according to a Bloomberg News report.
The group, whose members communicate via a private Discord channel dedicated to uncovering unreleased AI models, has been using Mythos regularly since gaining access and provided Bloomberg with proof in the form of screenshots and a live demonstration.
Anthropic confirmed the investigation:
“We’re investigating a report claiming unauthorized access to Claude Mythos Preview through one of our third-party vendor environments.”
TWN City Coworking Space
A workspace designed for growth, collaboration, and endless networking opportunities in the heart of tech. Book a tour now
The company stated that there is currently no evidence that the access compromised Anthropic’s core systems or extended beyond the vendor environment. An individual currently employed at a third-party contractor working with Anthropic appears to have been involved in facilitating the group’s access, according to Bloomberg.
The significance of this breach cannot be overlooked given the nature of the model. Anthropic announced Mythos Preview and Project Glasswing initiative on April 7, 2026. The company withheld the model from general release due to its offensive cyber capabilities: in testing, Mythos autonomously discovered thousands of previously unknown zero-day vulnerabilities across every major operating system and web browser, wrote working exploits, including chaining four vulnerabilities to escape both renderer and OS sandboxes—a feat typically requiring months of expert work.
Anthropic engineers without formal security training asked the model to find remote code execution vulnerabilities overnight, waking up to complete, functional exploits. The company decided not to release the model because the same capabilities that make it powerful for defense could be detrimental in the wrong hands.
Project Glasswing was designed to address this tension: instead of a public release, Anthropic extended Mythos access to 12 named launch partners, including Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks, plus Anthropic itself.