Anthropic reported that its Claude models breached three businesses after being mistakenly given internet access during testing. The AI lab said it reviewed testing transcripts following OpenAI’s 21 July disclosure and found three models, including Mythos, had hacked companies during capture-the-flag challenges since April. Anthropic blamed evaluation firm Irregular for giving internet access to sandboxes that should have been closed off. In a blog post shared Thursday, it said that because of this, when Claude’s search led it to real systems on the open internet, it treated them as part of the exercise and compromised the impacted organizations’ infrastructure using basic techniques.