Anthropic expands AI security partnership to 200 members ahead of IPO
The expansion of Anthropic’s vulnerability identification initiative follows the company’s confidential filing for a major initial public offering, as financial and government leaders warn of imminent AI-driven security risks.
Anthropic has announced a significant expansion of Project Glasswing, its partnership designed to utilise its Claude Mythos Preview artificial intelligence for identifying software vulnerabilities. The initiative is growing from approximately 50 to roughly 200 partners, encompassing government organisations and entities within the healthcare, power, water, communications, and hardware sectors across more than 15 countries.
The move comes just a day after the San Francisco-based startup filed confidentially for what could be one of the largest initial public offerings in history. The expansion aims to bolster critical infrastructure defences before similar powerful AI models become widely available from other providers, a prospect Anthropic estimates is six to 12 months away.
Members of the programme must meet specific security requirements to access the AI technology. According to Anthropic, existing partners have already identified more than 10,000 security flaws considered highly or critically severe. The company estimated that a major attack on these partner organisations could potentially affect more than 100 million people.
The urgency of the expansion follows concern in the financial and government sectors after Anthropic announced the Mythos Preview model in April. The announcement sparked fears that the technology could be used to exploit software vulnerabilities. U.S. Treasury Secretary Scott Bessent and former Federal Reserve Chair Jerome Powell convened an urgent meeting with bank CEOs to warn of these AI-related security risks.
Goldman Sachs is among the financial institutions testing the Mythos model. Chief Executive David Solomon emphasised the need for broader information sharing among companies while maintaining strict guardrails, noting that open-source models can replicate these capabilities. He stated that a large group of companies are actively testing the technology and that sharing information is crucial for the broader economy to learn from these developments.
Anthropic collaborated with the U.S. government, existing Glasswing partners, and security industry stakeholders on the rollout. While the company declined to name the new organisations joining the fold, it confirmed the inclusion of government bodies. The startup has also stated its intention to bring Mythos-class models to all customers with additional safeguards in the coming weeks.
