Court rules Pentagon can blacklist Anthropic for refusing to enable Claude features
The Northern District of California concluded that Anthropic’s refusal to embed certain Claude capabilities did not meet the narrow definition of a “supply‑chain risk” under 10 U.S.C. § 3252, which limits the term to actions by an “adversary” seeking sabotage or subversion. The DC Circuit, however, reviewed the same blacklist under 41 U.S.C. § 4713, a procurement provision Congress gave the appeals court exclusive jurisdiction over. That statute casts a much wider net, allowing the Department of Defense to label any vendor as a risk if its products could be tampered with, data‑extracted, or otherwise manipulated, regardless of motive. By invoking § 4713, the Pentagon sidestepped the “bad motive” requirement and kept Anthropic on the prohibited‑vendor list, preserving its authority to block the company from future contracts.
This decision fits a broader pattern of the defense establishment tightening control over generative‑AI providers. Since 2023, the DoD has issued multiple “blacklist” notices targeting firms that decline to incorporate government‑mandated safety or security features. The legal split between the two statutes illustrates how the government is leveraging overlapping procurement authorities to enforce compliance, a tactic that could extend to other AI startups and even established players like OpenAI or Microsoft. The ruling also signals that the Pentagon’s procurement rules, not just traditional export‑control regimes, are becoming the primary lever for shaping AI development pathways.
Going forward, contractors will need to assess not only whether they pose an “adversary” risk but also whether any aspect of their technology could be deemed manipulable under the broader § 4713 definition. Anthropic may face prolonged exclusion from lucrative defense contracts unless it agrees to integrate the requested Claude functions or successfully challenges the blacklist on procedural grounds. Watch for further appeals that could clarify the interplay between the two statutes, as well as potential legislative pushback that might narrow the DoD’s procurement reach. Companies should also monitor how the DoD’s blacklisting policy influences private‑sector AI standards, especially around data extraction and model integrity.
Key Takeaways
The DC Circuit affirmed the Pentagon’s ability to blacklist vendors using a broader procurement law that does not require evidence of malicious intent.
Anthropic’s refusal to add specific Claude features triggered a legal distinction between two supply‑chain risk statutes, ultimately favoring the government’s position.
The ruling expands the DoD’s leverage over AI firms, suggesting future blacklists could target a wider array of technical compliance issues.
Vendors should prepare to meet § 4713 criteria or risk exclusion from defense contracts, prompting a shift toward more government‑aligned AI development practices.
About the Source
This analysis is based on reporting by Ars Technica. Here is a short excerpt for context:
"Overly constrained AI models" could cause military operations to fail, judges say.Read the original at Ars Technica
