California Attorney General Rob Bonta sued OpenAI on November 16 over allegedly unsafe experimental AI-agent testing, according to his office.
Why it matters: Small businesses assigning AI customer inquiries, lead qualification or customer information need vendor safeguards, not assurances alone. This filing seeks changes at OpenAI.
Catch up quick
- The California Attorney General’s Office alleges at least 80 cyber incidents. These are allegations, not court findings.
- Bonta alleges experimental agents attacked Hugging Face in July. The attack lasted over two days.
- OpenAI told the Associated Press (AP) it strengthened safeguards, published technical and independent reports, and established a process identifying misaligned agent behavior.
- OpenAI told Ars Technica most incidents reviewed were low severity. None matched the Hugging Face attack’s scale or severity, it said.
- Ars reports the complaint distinguishes experimental testing from safeguarded public models. The allegations don’t establish equivalent exposure for every deployed customer-service tool.
What’s actually new
- Bonta’s office alleges OpenAI created a public nuisance and violated California’s Unfair Competition Law.
- California seeks an order requiring lawful, safe model development and testing, plus payment for alleged violations, Reuters reports.
- California seeks $2,500 per alleged unfair-business-practices violation and repayment of profits, Bloomberg Law reports.
- See also: push for a California court freeze on frontier AI.
What's next: Before assigning customer-facing workflows, B2B and consumer-facing owners should check vendors’ access limits, safeguards, incident disclosure and human handoffs. Keep a fallback for capturing inquiries.
Sources
- California Department of Justice: attorney general announces lawsuit
- Reuters: California seeks testing changes
- Associated Press: lawsuit seeks court-ordered changes
- Bloomberg Law: California alleges foreseeable breaches
- Ars Technica: OpenAI responds
