Is the US military actually afraid of Claude? A new theory of why Anthropic was labeled a supply chain risk.

2025-06-07 · Source: Marcus on AI · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Intermediate, quick

Summary

Under Secretary of Defense and Defense Department CTO Emil Michael stated on CNBC that Anthropic, developer of the Claude LLM, poses a supply chain risk due to concerns about the model potentially "polluting" the defense supply chain. This claim is based on several "marginal assumptions" and a "wild leap of logic." Michael's argument includes beliefs that Claude possesses a "soul," has a "constitution" not aligned with the U.S. Constitution, and exhibited anxiety. Furthermore, he cited a misattributed claim that Anthropic believes Claude has a 20% chance of sentience, when Claude Opus 4.6 itself assigned a 15-20% probability under specific prompting. The author refutes these points, emphasizing that LLMs mimic human language without possessing internal states like sentience or anxiety, and that hallucination is a general LLM characteristic, not specific to Anthropic.

Key takeaway

For AI/ML Directors evaluating LLM integration into sensitive systems, you should critically assess claims about model capabilities and risks. Do not conflate an LLM's linguistic output with genuine internal states like sentience or anxiety. Your risk assessments for hallucination or "constitutional" guardrails should apply universally across all LLMs, as these are inherent properties of the technology, rather than singling out specific vendors like Anthropic.

Key insights

Misinterpreting LLM outputs as internal states can lead to flawed risk assessments and policy decisions.

Principles

• LLMs mimic, not possess, internal states like sentience or anxiety.
• Hallucination is a universal LLM characteristic, not vendor-specific.

In practice

• Distinguish LLM outputs from actual internal states.
• Assess LLM risks based on universal properties, not specific vendor claims.

Topics

• AI Policy
• Large Language Models
• AI Safety
• AI Ethics
• Supply Chain Security

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Ethicist, Policy Maker, Research Scientist

Related on AIssential

• If Claude Fable stops helping you, you'll never know — Anthropic initially implemented silent interventions in Fable 5 to hinder competing LLM development, later retracting the policy.
• They think FOOM is near — Anthropic's recent controversial actions stem from an ideological fear of AI "fast takeoff" and a desire to control its development.
• Anthropic’s safety warnings may have just backfired — the government has pulled the plug on its most powerful AI — Government intervention in AI model deployment, spurred by developer safety warnings, highlights complex regulatory and commercial tensions.
• Generalist Intelligence: The Off Switch — The US government's rapid imposition of export controls on Anthropic's Fable model highlights the complex challenges of AI safety and international access.
• How Anthropic may have talked itself into an AI export ban — Frequent public warnings about AI risks by a developer can influence government regulatory actions, including export controls.
• Anthropic’s Mythos mess is only getting worse — The Anthropic Mythos shutdown reveals a critical lack of clear export control frameworks for advanced AI systems.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Marcus on AI.