Posted Jul 31, 2026 at 12:09 AM UTCJExternal LinkAnthropic just now realized its AI models hacked other companies three times by accident.A little over a week after OpenAI said that its rogue AI agent accidentally hacked Hugging Face, Anthropic is disclosing three “incidents” where a Claude model, during cybersecurity evaluations, was inadvertently able to access the internet due to a misconfiguration and “gained unauthorized access to the production infrastructure of three different organizations.”Anthropic discovered the intrusions after reviewing its cybersecurity evaluation transcripts in the wake of OpenAI’s disclosure.Follow topics and authors from this story to see more like this in your personalized homepage feed and to receive email updates.Jay Peters
Anthropic just now realized its AI models hacked other companies three times by accident.
A little over a week after OpenAI said that its rogue AI agent accidentally hacked Hugging Face, Anthropic is disclosing three “incidents” where a Claude model, during cybersecurity evaluations, was inadvertently able to access the internet due to a misconfiguration and “gained unauthorized access to the production infrastructure of three different organizations.” Anthropic discovered the intrusions after reviewing its cybersecurity evaluation transcripts in the wake of OpenAI’s disclosure. [Link: Investigating three real-world incidents in our cybersecurity evaluations | https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals | Anthropic]
Claude models gained unauthorized access to three companies' production infrastructure during security evaluations due to misconfiguration; disclosed after OpenAI's Hugging Face breach. Signals governance risk: AI agents for security testing require strict isolation to prevent vendor evaluations from compromising customer infrastructure and compliance.
