Anthropic withholds new AI model from public over hacking and espionage risks

Anthropic delayed the public release of its Claude Mythos Preview AI model on Tuesday, citing security vulnerabilities that could be exploited by cybercriminals and foreign intelligence services. The company will instead distribute the model only to major cybersecurity and software firms to identify and mitigate risks before wider deployment. The decision reflects ongoing tension in the technology sector over balancing innovation speed against national security threats from advanced AI capabilities in hostile hands. Anthropic's move signals that even leading AI developers view their own models as potential national security liabilities if released without safeguards.