Anthropic says its AI model Mythos should not be released to the general public because of its capability to find software and computer-system vulnerabilities. The company argues that making such a tool widely available could help malicious actors steal data or disrupt critical infrastructure. Both outlets report that Anthropic has taken a staged approach: Mythos is provided to selected organizations rather than offered broadly.
According to The Next Web, as of early June Anthropic expands Mythos access to 150 additional organizations, bringing the total to about 200 organizations operating across 15 countries. Bloomberg similarly describes Mythos as a tool designed to identify vulnerabilities, and states that Anthropic initially limited access to a small set of carefully chosen parties.
Together, the accounts indicate that Anthropic is simultaneously emphasizing the risks of public release while increasing the number of vetted users. The company’s position centers on controlling distribution to reduce the likelihood that the model’s security-focused strengths are misused.
