IC Technology & Surveillance — 2026-04-09
IC Agencies Briefed on Anthropic AI Model That Identifies Thousands of Software Vulnerabilities
Anthropic unveiled Project Glasswing, granting restricted access to Claude Mythos Preview, an unreleased AI model that has identified thousands of high-severity vulnerabilities in every major operating system and web browser, to AWS, Apple, Cisco, Google, and Microsoft. NSA, CISA, and multiple Defense Department components received briefings on the model's full capabilities before the external announcement. Former NSA executive director Morgan Adamski warned of a coming equity conversation over whether discovered vulnerabilities should be disclosed to vendors or retained for offensive use. The Pentagon previously designated Anthropic a supply chain risk, complicating any intelligence-sharing arrangement.
The vulnerabilities-equities dilemma is the central IC implication. If Mythos Preview can discover thousands of high-severity flaws across all major operating systems, the intelligence value of retaining undisclosed vulnerabilities for offensive use collides directly with the defensive imperative to patch them. Senator Warner's concern about discovery outpacing patching suggests the IC may face congressional pressure to disclose more rapidly than offensive equities allow. The Pentagon's earlier designation of Anthropic as a supply chain risk adds friction to any intelligence-sharing arrangement.
Sources: Defense One · Nextgov/FCW