Episode

Who Decides AI Is Safe Enough? | Sept 29, 2026

Name: Who Decides AI Is Safe Enough? | Sept 29, 2026
Uploaded: 2026-09-30T03:00:04+00:00
Duration: 13 min 46 s
Description: Who should decide whether an AI system is ready for release? Sept 28–29, 2026 archive: court requests, reviews and risk disclosures are qualified below.

September 30, 2026 13:46 12 views Watch on YouTube

Read the full story

Singapore Sets Phased AI Risk Guidelines for Financial Institutions

Who should decide whether an AI system is ready for release?

Sept 28–29, 2026 archive: court requests, reviews and risk disclosures are qualified below.

How do we get the benefits of AI without losing control of it? Six AI ethics stories from Sept 28-29, 2026 on AI copyright, AI safety, AI agents, AI regulation and AI governance - from OpenAI, Anthropic and NVIDIA AI to a U.S. appeals court, Australia's banks and the United Nations. Each story carries its status: a court ruling, a request, an announcement, a review, a risk disclosure or a confirmed fact.

1. Who owns AI training data? COURT RULING - FURTHER REVIEW POSSIBLE. On Sept 29, 2026 a federal appeals court upheld a ruling that Ross Intelligence's use of Thomson Reuters' Westlaw material to train its AI legal-research tool was not fair use. The written opinion is still sealed. Ross built a search tool, not generative AI - this does not mean every AI model trained on copyrighted material is illegal.

2. Florida vs. OpenAI: COURT REQUEST - NOT A RULING. On Sept 28 Florida's attorney general asked a state court for six restrictions on OpenAI, including no new AI models without an outside safety check and no ChatGPT for Florida minors. No judge has ruled. These are not final rules.

3. NVIDIA and AI agents: COMPANY ANNOUNCEMENT. On Sept 28 NVIDIA announced its Open Agent Safety Platform. OpenShell, open-source software that runs an agent in a sandbox and limits what it can touch, is out now. Sentry, a hardware watchdog that can stop an agent that leaves its boundary, is still a design.

4. AI and your loan: REGULATOR REVIEW. Reuters reports that Australia's securities regulator, ASIC, has told banks it will review how they use AI with customers, including lending decisions. A review, not a new rule. The loan example in the video is James's own illustration.

5. Anthropic: RISK DISCLOSURE - NOT A PREDICTION. Reuters reviewed Anthropic's draft IPO papers, which are not yet public. The risk factors list things that COULD go wrong, including models that might "resist shutdown" or "conceal or manipulate information." Risk factors, not forecasts.

6. AI at the United Nations: CONFIRMED FACT. At the UN General Assembly's general debate (Sept 22-28), 128 delegations raised AI; four years ago, five did. Talks, not rules - no UN resolution on AI was adopted this week.

THE RESPONSIBLE AI CHECKLIST: test it, protect private data, keep humans involved in important decisions, know where training data comes from, control what AI agents can access, and make sure someone is responsible when something goes wrong.

We are pro-AI. Good AI ethics should not stop innovation - it should help people trust innovation.

Today's question: Who should decide when AI systems are safe enough to release? The company that built it? The government? An independent safety group? A court? Or some combination of them all? Put your answer in the comments.

CHAPTERS

How do we get the benefits of AI without losing control?

Welcome: AI is moving fast

1 - Who owns the data used to train AI? (Thomson Reuters v. Ross)

2 - Florida asks a court for new OpenAI safety controls (not final)

3 - NVIDIA: a security layer around AI agents

4 - When AI helps decide if you get a loan (Australia's ASIC reviews banks' AI)

5 - Anthropic's warning to investors: a risk disclosure, not a prediction

6 - AI at the United Nations: who makes the rules?

What all six stories have in common

My takeaway: pro-AI, with limits

Your question: who should decide when AI is safe enough?