Trump Said the Government Will No Longer Use Anthropic’s A.I.

Source: The New York Times

Original link: https://www.nytimes.com/2026/02/27/briefing/trump-pentagon-anthropic-ai.html

Pulse — Trump Said the Government Will No Longer Use Anthropic’s A.I.

Source: The New York Times

Link: https://www.nytimes.com/2026/02/27/briefing/trump-pentagon-anthropic-ai.html

Pulse (AI Failure): U.S. Government Ceases Use of Anthropic AI Following Undisclosed Issues

The What:

The U.S. government, under former President Trump’s administration, announced it will no longer utilize Anthropic’s AI systems. Specific technical reasons for this operational halt remain undisclosed in the source excerpt. Unknown from source excerpt: whether the cessation was due to model reliability failures, safety filter bypass, unauthorized data handling, or other operational breakdowns. Verification needed on the exact malfunction or governance breach.

The Why (Governance Gap):

The decision suggests a governance gap potentially rooted in insufficient accountability or failure to maintain robust human-in-the-loop oversight. The absence of transparent incident reporting and risk management protocols indicates an unmanaged AI risk appetite, undermining trust and operational continuity. The low governance meta-score (0/100) further implies inadequate governance frameworks or enforcement mechanisms were in place.

The How (Frameworks & Laws):

Under the EU AI Act, Anthropic’s AI would likely be classified as high-risk if deployed in government decision-making, triggering strict obligations for risk management, transparency, and human oversight—noncompliance could justify discontinuation. The NIST AI Risk Management Framework (AI RMF) mandates continuous GOVERN, MAP, MEASURE, and MANAGE cycles to ensure AI system reliability and safety; failure to implement these could lead to operational withdrawal. ISO/IEC 42001 standards require comprehensive AI impact assessments and controls (AIMS), which if absent, would constitute a critical governance failure.

System Design (Prevention):

To prevent similar failures, a technical architecture should incorporate Retrieval-Augmented Generation (RAG) with verified Golden Datasets to ensure factual grounding. Runtime monitoring must detect model drift and bias in real-time, triggering refusal mechanisms when confidence thresholds are not met. Sandboxed execution environments for agentic AI components would contain potential unsafe behaviors. Compliance with ISO/IEC 42001’s AIMS controls and continuous NIST AI RMF governance cycles would institutionalize oversight and risk appetite management, ensuring operational resilience and regulatory alignment.