Anthropic's Powerful AI Model, Claude Mythos, Causes Security Concerns and Promotes AI Partnership with Broadcom and Google

Anthropic unveiled Claude Mythos Preview, a frontier AI model with cybersecurity capabilities so powerful the company decided not to release it publicly. In testing, Mythos identified thousands of zero-day vulnerabilities across every major operating system and web browser — including flaws believed to be decades old that had evaded repeated human-run security audits. Anthropic concluded the model's ability to autonomously find and exploit software vulnerabilities surpasses all but the most elite human security researchers, making unrestricted public access too risky.

Instead of a public launch, Anthropic announced Project Glasswing — a controlled initiative giving early access to a select group of organizations including AMZN AWS, Apple, AVGO Broadcom, Cisco, CRWD CrowdStrike, Google, JPMorgan Chase, Microsoft, NVDA NVIDIA, and Palo Alto Networks. The mission: use Mythos's vulnerability-finding capabilities offensively on behalf of defenders, patching critical infrastructure before bad actors can exploit the same weaknesses. AVGO shares rose 5% on news of an expanded long-term chip deal accompanying the Glasswing announcement.

The dual-use nature of Mythos represents a watershed moment for AI and cybersecurity. The model's capabilities make it a potential force multiplier for defenders at scale — but also an unprecedented threat if accessed by state-sponsored hackers or criminal organizations. Anthropic's controlled release model sets an important precedent for how frontier labs will navigate the tension between commercial deployment and responsible AI development as models gain capabilities that outpace existing regulatory frameworks.