Expanding the Cyber Verification Program

My take: Anthropic expanded its Cyber Verification Program and organized it into three access tiers for security teams: one for day to day defensive operations, another for authorized penetration testing, and a more limited one for critical systems like power grids or flight control. The core idea is that security professionals get access to Anthropic's most capable models, with fewer restrictions, specifically to find vulnerabilities before attackers do.

According to Anthropic, partners in this program (previously known as Project Glasswing) found more than 129,000 verified vulnerabilities between April and July of this year, and its own open source efforts found 5,500 more; over 33,000 of those flaws were rated critical or high severity. These are figures reported by the company itself about its own tool, so they are worth reading with your own judgment while waiting for independent confirmation over time, but the underlying trend, AI speeding up security work, shows up across the industry.

For any business, this is a clear signal of where cybersecurity is heading: AI is not just a risk anymore (deepfakes, automated attacks), it is also becoming part of the defense. If you have a technical team or rely on outside vendors to protect your systems, it is worth asking what AI tools they are using to find flaws before someone else exploits them.

Do you know whether the company protecting your systems, yours or a vendor's, is already using AI to hunt for vulnerabilities, or still relies only on manual reviews?