OpenAI says more than 15,000 accounts tried to extract its models' reasoning, and links a core group to people associated with Moonshot AI, maker of Kimi. Anthropic made a similar claim in February.

By OpenAI's own account, no encryption was cracked and no database breached. The charge is breaking its terms of service.

Both labs know that charge from the other side. Reddit is suing Anthropic for breaching its user agreement by scraping posts to train Claude. Canadian news publishers have sued OpenAI for, among other things, breaching their websites' terms.