Software: Anthropic's Project Glasswing news, which we see as a positive for cyber consolidators

Research evidence excerpt

Software: Anthropic's Project Glasswing news, which we see as a positive for cyber consolidators

RBC Capital Markets, LLC

Matthew Hedberg (Head of

Global TIMT Research)

(612) 313-1293,

matthew.hedberg@rbccm.com

Dan Bergstrom (Analyst)

(612) 313-1254,

dan.bergstrom@rbccm.com

May 22, 2026 Michael Richards (AVP)

(332) 204-5767,

michael.richards@rbccm.com Software: Anthropic's Project Glasswing news, Matthew Swanson (Analyst)

(612) 313-1237,RESEARCH which we see as a positive for cyber consolidators matthew.swanson@rbccm.com

Simran Biswal (AVP)

Our view: Anthropic's Project Glasswing, a partnership with ~50 organizations using its unreleased (646) 618-6894, simran.biswal@rbccm.com Claude Mythos Preview model, has uncovered >10,000 high/critical-severity software vulnerabilities Rishi Jaluria (Analyst)

in its first month, accelerating bug discovery by 10x+ for partners including NET, PANW, MSFT, ORCL (415) 633-8798,

and CSCO (which were all named in the blog post). Anthropic noted the bottleneck in cybersecurity rishi.jaluria@rbccm.com

has shifted from finding vulnerabilities to patching them, creating both near-term risk and a long-term

opportunity for cybersecurity vendors. Mythos Preview achieved a 90.6% true-positive rate on triagedEQUITY open-source findings, with precision that partners describe as better than human testers. Anthropic is

already commercializing adjacent capabilities through Claude Security (in public beta), which has been

used to patch >2,100 vulnerabilities in three weeks. The company is also releasing its scanning harness,

threat model builder, and partner-developed tooling to qualifying security teams. Anthropic plans to

release Mythos-class models publicly once stronger safeguards are developed, while acknowledging