The Artificial Intelligence Show #201: Anthropic vs. Pentagon Round 2, AI Job Impact Study, Services as the New Software & GPT-5.4
351 snips
Mar 10, 2026 A tense tech-policy clash resurfaces as a government blacklisting collides with continued military use of AI. New research claims AI can theoretically handle 94% of knowledge work and sparks debate about job exposure. A venture bet suggests services, not just software, are ripe for automation. Plus a major model release beats professionals on benchmarks and fresh security, copyright, and privacy controversies heat up.
AI Snips
Chapters
Transcript
Episode notes
Anthropic Blacklist Creates Operational Contradictions
- Anthropic was designated a supply chain risk by the Pentagon, prompting Treasury, State, and HHS to end use while the military still uses Claude in active operations.
- The designation makes Anthropic the first U.S. firm treated like a foreign adversary, creating legal and operational contradictions.
Assess Jobs By Tasks Not Titles
- Break jobs into task-level components using resources like O*NET and then map AI exposure per task to forecast disruption.
- Paul recommends tools like JobsGPT to assess exposure and project how model improvements change role risk.
Create Focused Company Evals For Model Releases
- Build company-specific evals to test model performance on the three to five outcomes that matter most to your business.
- Anthropic and Paul recommend task-level benchmarks rather than generic IQ tests to track exposure and risk over time.