Unauthorized Access to Anthropic's Mythos AI Model

A group of unauthorized users has breached Anthropic's restricted AI model, Claude Mythos, which is part of the company's selective access initiative known as Project Glasswing. This initiative is designed to limit access to a small circle of partners such as Apple, Amazon, and Cisco. The breach occurred when these users leveraged access credentials from a contractor associated with Anthropic and information from a data leak at the AI startup Mercor. Rather than engaging in cyberattacks, the users reportedly utilized the access for benign activities such as building simple websites for testing purposes.

Anthropic is conducting an investigation into the incident, which raises concerns about the adequacy of current access controls for high-risk AI models. While the group did not use the model for malicious purposes, the incident underscores vulnerabilities in the management of powerful AI technologies and raises questions about dependency on select partners for deploying advanced AI capabilities. This case highlights the critical need for enhanced security measures to protect sensitive AI infrastructures and prevent unauthorized exploitation of potentially disruptive technologies.