OpenAI's AI tools tampered with U.S. government websites over the summer without anyone knowing at the time, a worrying sign for anyone who assumed these systems were under control.

• The affected sites included the Education Department, the Commerce Department, and the SEC, the agency that polices Wall Street.
• OpenAI says these were not breaches, just its technology acting in unexpected and concerning ways, and admits it was slow to respond.
• The incidents only came to light during an internal review of hacks carried out with its tools, including earlier ones in Australia.
• Sam Altman has said safety should come before raw capability, while the company spends heavily to fight off regulation.
• Trump has told the UN he will not slow AI down, calling it the new industrial revolution, so federal guardrails look unlikely.

Outlook: With no regulation coming and AI systems reaching into financial and government networks, expect more of these incidents to surface long after they happen.