AI evaluation lab Irregular's report on its role in hacking incidents involving OpenAI, Anthropic, and Meta models faces criticism over unanswered questions (Alexander Martin/The Record)

Irregular’s report on its involvement in hacking incidents targeting OpenAI, Anthropic, and Meta models has drawn criticism for leaving key questions unanswered. In response to the Hugging Face breach and evidence that Astra reached a critical cyber threshold, OpenAI halted reinforcement-learning training for two weeks and revised its safety protocols. The controversy highlights ongoing concerns about AI model security and accountability.