Reported by 1 outlet
Anthropic Engineers Embedded at NSA — Mythos Model Deployed for Offensive Cyber Operations, DoD "Supply Chain Risk" Paradox
Why this score
- Breadth
- 0.50 / 2.0 · 1 source
- Authority
- 1.00 / 2.0 · best outlet weight
- Severity
- 1.30 / 5.0 · CVSS + content signals
- Velocity
- 0.00 / 2.0 · updates in 48h
- Pickup speed
- 0.00 / 0.6 · time to second source
- Recency
- × 0.00 temporal multiplier
- Community
- + 0.00 / 1.5 · reddit signal
- Score
- 0.0 / 10 = base × recency + community
Timeline
-
original reporting
Anthropic Engineers Embedded at NSA — Mythos Model Deployed for Offensive Cyber Operations, DoD "Supply Chain Risk" Paradox
the-decoder.com · 129d ago · June 7th, 2026
Anthropic has placed approximately six "forward-deployed" engineers inside the National Security Agency to customize and deploy its most capable AI model, Mythos Preview, for offensive cyber operations — including potential network infiltration against China and Iran, per the Financial Times. The arrangement runs concurrently with Anthropic's active lawsuit against the Department of Defense, which designated the company a "supply chain risk" in March 2026 after Anthropic refused to remove guardrails on Claude for mass domestic surveillance and autonomous weapons.
What Mythos can do: Anthropic's red team confirmed Mythos can identify and exploit zero-day vulnerabilities in every major operating system and web browser. The model uses an agentic architecture that autonomously reads source code, forms vulnerability hypotheses, writes and executes test cases, and iteratively confirms exploitable bugs without human guidance at each step. A complete exploit pipeline for a complex target costs under $2,000 and completes in under 24 hours. Britain's AI Security Institute measured a 73% success rate on expert-level vulnerability research tasks.
The paradox: The Pentagon ordered a complete purge of Anthropic technology from DoD systems by August 2026, citing supply-chain risk — yet the NSA (a DoD agency) secured an explicit carve-out for Mythos operations. Anthropic balked at uses carrying domestic reputational and legal cost (mass surveillance of Americans, autonomous weapons) while staffing a deployment aimed outward at foreign networks. The Electronic Frontier Foundation and congressional oversight figures have warned the arrangement bypasses democratic checks.
Project Glasswing expansion: On June 2, Anthropic expanded Mythos access to ~150 organizations across 15+ countries, up from 50 mostly-US partners in April. Named members include Okta, Samsung, NATO, and ENISA. Partners have surfaced over 10,000 high- or critical-severity software vulnerabilities; an internal Anthropic scan of 1,000 open-source projects flagged 23,019 potential vulnerabilities, 6,202 estimated as high or critical. The program now covers power, water, healthcare, and communications infrastructure sectors. Anthropic filed confidentially for an IPO at a valuation near $1 trillion.
Bottom line: The nation's most capable offensive AI cyber tool is operational on classified networks, staffed by private-sector engineers, deployed by an agency whose parent department considers their employer a security threat, and is simultaneously being commercialized to 150 organizations globally — without any public congressional hearing or legislative framework governing its operation.