Anthropic opens its most powerful AI models to more security teams | Reuters
Anthropic is widening a program that lets vetted cybersecurity professionals test its strongest AI models with fewer safeguards. The company says its Project Glasswing initiative has already helped find over 100,000 software vulnerabilities this year, and that the real number is likely at least five times higher.
- Project Glasswing has helped uncover more than 100,000 software vulnerabilities this year.
- Anthropic thinks that figure is an undercount, possibly by five times or more, because it comes from a survey of only a limited number of partners.
- The expansion gives more vetted security teams access to the company's most capable models with reduced safeguards.
Anthropic is widening access to a program that lets vetted cybersecurity professionals test its most capable models with fewer guardrails than the public gets. The program is called Project Glasswing, and according to the piece, it has helped uncover more than 100,000 software vulnerabilities this year. That number comes from a survey of a limited set of partners, which is why Anthropic says it is almost certainly an undercount and expects the real impact to be at least five times larger.
The reason this matters is the tension it exposes. The same capabilities that let a model find bugs in code are the ones that could help someone exploit them, so Anthropic is effectively betting that giving defenders earlier and deeper access will outpace attackers who would get there anyway. The report frames the expansion as a bid to get those tools into the hands of security teams before they are commonly available, but it leaves open how Anthropic decides who qualifies as "vetted" and what safeguards remain in place when the guardrails are loosened.
The lack of detail on those points is the main gap. The excerpt gives scale but not method: no breakdown of vulnerability severity, no information on how many organizations are involved, and no independent verification of the 100,000 figure. Treat the headline number as the company's own estimate from a partial sample until more data surfaces.