Anthropic limits access to Mythos, its new cybersecurity AI model (arstechnica.com)

Anthropic has introduced its new cybersecurity AI model, Claude Mythos Preview, but is limiting access to a select group of vetted organizations, including major tech firms like Amazon, Apple, and Microsoft. This move comes in the wake of two significant data leaks from the company, which raised concerns about its security practices and data vulnerabilities. Anthropic is also in discussions with the US government regarding the model's application, underscoring its potential importance in national security. Mythos stands out as a "general purpose" model with capabilities to identify cyber vulnerabilities on an unprecedented scale, while also presenting risks as it could be used to exploit these weaknesses if it falls into the wrong hands. Anthropic's cautious approach reflects the dual-edged nature of such powerful AI technologies, as stated by Dianne Na Penn, the company’s head of product management. By restricting access, Anthropic aims to ensure that select organizations can leverage these advanced security capabilities to better protect against cyber threats, while also recognizing the potential for misuse.