Report: Discord Group Uses Claude's Supposedly Secret Mythos

Artificial Intelligence & Machine Learning , Next-Generation Technologies & Secure Development , The Future of AI & Cybersecurity

An unauthorized group of users gained access to Claude Mythos Preview artificial intelligence model and have regularly used it since the day that AI firm Anthropic revealed the model's existence while pronouncing it too dangerous to release to the public, reports Bloomberg.

See Also: Context Drives Security in Agentic AI Era

Anthropic made a splash when earlier this month it said it would reserve access to a select group of companies joined together under "Project Glasswing," with the understanding that they would use the model to find and fix security vulnerabilities before hackers get access to equally powerful tech. Members include Nvidia, Apple, Amazon and Cisco (see: Anthropic Calls Its New Model Too Dangerous to Release).

A source told Bloomberg the unauthorized users belong to a private Discord channel dedicated to unreleased models. An apparent member of the group told the newswire that users have not used Mythos to hunt for new exploits.

Anthropic has touted the vulnerability finding properties of Mythos in a publicity campaign that has received some outside validation, as from the AI Security Institute in Great Britain, which found the model to be "a step up over previous frontier models."

The source told Bloomberg the Discord group deployed a mix of tactics to access Mythos, including using access the source has as a third-party contractor for Anthropic. The group "made an educated guess about the model's online location based on knowledge about the format Anthropic has used for other models." The person said such data was leaked in a recent breach at AI startup Mercor (see: Mercor Breach Linked to LiteLLM Supply-Chain Attack).

An Anthropic spokesperson told the newswire that it is investigating the matter but that it has no evidence of unauthorized Mythos use beyond the third party's IT environment. The source told Bloomberg the Discord group also has access to other unreleased Anthropic models.

Anthropic's release of Mythos to select partners received a rejoinder from rival firm OpenAI, which days later released GPT‑5.4‑Cyber with the stated intention of making it "as widely available as possible." OpenAI said it will rely on user identity verification and "trust signals" to safeguard its vulnerability-seeking AI model from being put to bad uses (see: OpenAI Touts Wider Access to Its New Cyber Model).