Anthropic Engineers Support NSA Offensive Cyber Operations

The Financial Times reported that Anthropic has deployed around half a dozen engineers to the US National Security Agency to help the agency use Mythos, Anthropic's frontier cybersecurity model, which the company has declined to release publicly. According to the FT, the staff work in a "forward-deployed" capacity, and one person said the model would be useful for infiltrating networks in nations such as China or Iran. Summarizing the FT, TechCrunch emphasized an important caveat: it is unclear whether the engineers or Mythos are yet being used in the agency's active hacking operations. The NSA declined to confirm or deny the reporting, and Anthropic did not comment to TechCrunch.

Axios first reported in April that the NSA was using or seeking Mythos despite a federal restriction on Anthropic's technology. That restriction followed a Defense Department move to designate Anthropic a "supply-chain risk" after the company sought to limit use of its models for mass domestic surveillance and autonomous weapons. The current arrangement is notable because it proceeds even as Anthropic and the Pentagon remain at odds over how its models may be used.

The reporting landed the same week Anthropic published a post titled "When AI Builds Itself," in which it warned that frontier models are moving closer to recursive self-improvement, the ability of systems to improve their own capabilities by writing their own code, and called for a coordinated plan to pause frontier development if society cannot manage the pace. Anthropic said it plans to convene policymakers, researchers, and civil society in the coming months. Outlets including Decrypt and The Guardian highlighted the contrast between a lab publicly urging caution and the reported use of its tooling for government offensive-cyber preparation.

Models built to automate vulnerability discovery and exploit generation blur the line between defensive and offensive use, which is why access is tightly held. As an industry pattern, exposing such capabilities through third-party contractors or embedded vendor staff raises supply-chain and access-control risk. That risk is not hypothetical here: Anthropic has said it is investigating an earlier report that unauthorized users gained access to a Mythos preview through one of its third-party vendor environments, an incident reported in April by outlets including TechCrunch, Bloomberg, and CBS News.

• •Whether congressional or regulatory bodies cite the FT and Axios reporting and seek audit logs or contractual disclosures.
• •Any official clarification from the NSA or Defense Department on the scope of embedded staff and whether they customize models or take part in live operations.
• •The makeup and output of any convenings that follow Anthropic's pause proposal, and whether other frontier labs engage.
• •Further disclosures about the earlier unauthorized-access incident and the vendor environment involved.