Anthropic expands powerful Mythos AI access to 200 security partners
Anthropic announced on Tuesday that it is expanding access to its powerful, vulnerability-hunting artificial intelligence by quadrupling the size of its elite cybersecurity collective, Project Glasswing.
The San Francisco-based startup is scaling the initiative from 50 organizations to roughly 200 vetting partners. Each new member must clear stringent security requirements before gaining access to the highly restricted Claude Mythos Preview. According to Anthropic, the new influx of Glasswing partners spans more than 15 countries and is aimed directly at hardening digital defenses across vital global sectors, including healthcare, power grids, water systems, telecommunications, hardware manufacturing, and various government agencies, News.Az reports, citing Reuters.
The expansion was coordinated in close collaboration with the U.S. government and established security leaders. It comes at an incredibly high-stakes moment for the company: just a day prior, Anthropic filed confidentially for what Wall Street expects could be one of the largest initial public offerings (IPOs) in history.
When the Mythos Preview was initially unveiled in April, it triggered widespread panic throughout public and private sectors. The AI boasts an unprecedented ability to sniff out minor software bugs and autonomously figure out how to stitch them together into devastating, weaponized exploits. The threat was deemed so acute that U.S. Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell convened an emergency meeting with banking CEOs to warn of systemic risks to the financial sector, prompting an industry-wide scramble to patch vulnerabilities.
Anthropic revealed that its initial batch of partners has already utilized the model to uncover more than 10,000 "highly or critically severe" software flaws. While some independent cybersecurity experts argue that the fears of widespread, AI-driven hacking are somewhat overblown, Anthropic's internal calculations suggest that a major successful cyberattack on its partner organizations could realistically impact up to 100 million people.
The startup emphasized that the core mission of Project Glasswing is to aggressively shore up critical global infrastructure before AI models of this caliber become widely accessible from competing open-source or commercial vendors—a milestone Anthropic estimates is a mere 6 to 12 months away. Following this controlled defensive rollout, Anthropic plans to introduce Mythos-class models to its broader commercial customer base with integrated, hardcoded safety protocols in the coming weeks.
By Aysel Mammadzada
Similar news
