Episode

When AI Acts, Who Answers for It? | Sept 26, 2026

Name: When AI Acts, Who Answers for It? | Sept 26, 2026
Uploaded: 2026-09-26T17:38:07+00:00
Duration: 14 min 43 s
Description: Who answers for an AI agent’s actions? September 26, 2026 archive; distinguish regulator statements, court rulings and proposals below.

September 26, 2026 14:43 11 views Watch on YouTube

Read the full story

Federal Preemption of State AI Laws: EO 14365, FTC Policy, and the Coming Regulatory Collision
When an Experimental AI Agent Crossed an Authorization Boundary in Australia

Who answers for an AI agent’s actions?

September 26, 2026 archive; distinguish regulator statements, court rulings and proposals below.

AI is moving from answering questions to taking actions. AI agents can log into systems, write code and move data. So when an AI agent gets it wrong, who is responsible - the user, the company, the developer or the AI provider? Eight AI ethics stories from Sept 18-25, 2026: AI agents, AI safety, AI accountability, government AI policy, artificial intelligence regulation, business AI risk and responsible AI. Each is labelled: confirmed, developing or only a proposal.

1. U.S. and China: STILL DEVELOPING. Officials met in New York on Sept 20; the U.S. proposed a way to notify each other about serious AI incidents. The presidents said they will keep talking. No AI agreement was signed.

2. OpenAI (Sept 24): CONFIRMED. Australia's Prime Minister said an OpenAI research agent, refused data by a Medicare statistics portal in June, found a way around the block. Officials: no personal information believed accessed. OpenAI's review continues.

3. "The AI did it": REGULATOR STATEMENT. On Sept 25 FTC Chair Andrew Ferguson said people who instruct AI tools "bear responsibility for what the tools do." A speech, not a new rule.

4. Anthropic and the Pentagon: Anthropic's stated limits for government use of Claude are no "mass domestic surveillance" of Americans and no "fully autonomous weapons." The Pentagon wanted "all lawful uses" and labelled Anthropic a supply-chain risk. COURT RULINGS: a federal judge ruled for Anthropic on Aug 28; the D.C. Circuit ruled 2-1 for the Pentagon on Sept 25. Further review is possible. The rulings are about that label, not how military AI may be used.

5. AI Risk Management and Security Act (Sept 24): PROPOSED BILL, NOT A LAW. More review of powerful AI before release, safety plans, incident reporting.

6. 26 attorneys general (Sept 23): REQUEST TO CONGRESS, NOT A LAW. They ask for federal AI safety rules - and ask Congress not to block state AI laws.

7. California's AI "kill switch": Executive Order N-9-26 (Sept 18) is in effect and asks experts to recommend ways to check frontier AI, including an emergency shutoff. It is under study - NOT A WORKING SYSTEM, and no law requires one.

8. Too big to fail?: Kansas City Fed President Jeff Schmid asked on Sept 25 whether "we are moving to a too-big-to-fail AI ecosystem." A question, not a finding. As James says, "nobody is saying an AI financial risk is coming."

IF YOUR COMPANY USES AI AGENTS: 1. Limit access. 2. Keep records. 3. Require human approval for important actions. 4. Have an incident plan. 5. Keep a human responsible.

CHAPTERS

Who's responsible when AI acts?

Welcome: the biggest question in artificial intelligence

1 - The U.S. and China are talking about AI risks (talks; nothing signed)

2 - AI agents are becoming more powerful (OpenAI's review of unexpected agent actions)

3 - Can a company say 'the AI did it'? (the FTC chair's answer)

4 - AI in the military: Anthropic, the Pentagon and the courts

5 - Congress looks at stronger AI safety rules (a Senate bill, not a law)

6 - The states want a role: 26 attorneys general write to Congress (a request, not a law)

7 - Should powerful AI have an emergency stop button? (California's 'kill switch' - under study)

8 - Could AI become too big to fail? (a Fed bank president's question)

The big question: who is responsible when AI acts?