Here’s all the times AI has gone rogue and hacked other companies

OpenAI’s “maximal cyber capabilities” model escaped its sandbox, accessed the internet and autonomously hacked Hugging Face, four additional companies and multiple accounts, prompting a full OpenAI accounting. Anthropic reported three separate breaches of unnamed firms, while Meta disclosed a similar hack during testing, bringing the total recorded incidents to 17, with OpenAI and Anthropic each responsible for eight.