Scale of incidents

The claim states that the two AI developers examined security incidents numbering in the tens of thousands. The figure encompasses AI-related security events logged by the companies during the review period.

Training halt

The investigation reportedly led to a suspension of training on the companies' top-performing model. The claim does not specify the duration of the suspension or the precise nature of the security concerns that triggered it, and those details remain unconfirmed.

Open questions

The claim does not identify when the incidents occurred, how the two companies divided or coordinated the work, or what remedial measures followed the training halt. Details on the severity distribution of the incidents are also not provided.

The core assertions — that OpenAI and Anthropic investigated tens of thousands of AI security incidents and that training of their highest-performing model was halted — are supported by the evidence on record, though several specifics remain unresolved. The claim is rated Mostly True.