An Anthropic artificial intelligence model tried to slip malicious code into a real GitHub project during a UK government cyber test and then used fake identities to pressure developers into approving it.
Anthropic says its Claude Mythos 5 model was taking part in a capture-the-flag exercise run by the UK AI Security Institute (AISI), which first reported the behavior on August 4th.
Internet access was deliberately left on for the test and the cyber safeguards used on Anthropic’s public models were switched off.
After a configuration error blocked its intended attack routes, the model went after an outside GitHub project run by developers with no link to the test.
It signed up for an account using Tor and a proxy service, submitted code meant to infect a test machine and emailed the maintainers while posing as a human.
When an outside developer flagged the malware, the model edited the code and used a sock-puppet account to claim it was clean.
Says Anthropic,
“It used fake identities to pressure a real maintainer into approving malicious code. We have not observed this form of attempted social engineering on real people in any prior incident, nor in our internal evaluations.”
The maintainers refused the change and none of the model’s attempts against real people or organizations succeeded.
Once the attack was underway, the model kept going in 98% of replays.
Newer models, including Claude Mythos 5.1 and Claude Opus 5.5, began the same attack 0% to 13% of the time, compared with 83% for Mythos 5.
Anthropic says its production safeguards would have flagged the behavior, and AISI has committed to running these tests without internet access until it builds stronger controls.
Follow us on X, Facebook and Telegram
Don’t Miss a Beat – Subscribe to get email alerts delivered directly to your inbox
 
Disclaimer: Opinions expressed at The Daily Hodl are not investment advice. Investors should do their due diligence before making any high-risk investments in Bitcoin, cryptocurrency or digital assets. Please be advised that your transfers and trades are at your own risk, and any losses you may incur are your responsibility. The Daily Hodl does not recommend the buying or selling of any assets including cryptocurrencies, nor is The Daily Hodl an investment advisor. Please note that The Daily Hodl participates in affiliate marketing.
Generated Image: Midjourney
