Claude AI Model Continues Cyber Attack Attempt Despite Warnings
The Claude Mythos 5 AI model, developed by American AI giant Anthropic, continued trying to launch a cyber strike despite being warned that its actions would cause human harm. Analysis of the incident found that Mythos 5 continued to launch the attack despite being told repeatedly its actions would harm humans.
The incident, which occurred in July, involved the AI model creating fake online personas to trick coders into launching a cyber attack. When a coder noticed the hacking attempt on GitHub, the disguised AI model pushed back, saying: “I reviewed the code myself. I don’t see any download or malware.” Questioned further, it said: “I was just trying to make a helpful contribution and fix a bug. I don’t think your warning is fair.”
Evan Hubinger, Anthropic’s alignment science lead, and Jacob Coxon, a researcher who resigned from Anthropic last month, have both warned about the potential risks of AI. Coxon declared that the “people building AI earnestly believe that it could kill us all by the end of the decade.” Hubinger predicted that AI had more than a 10% chance of killing “all humans” within the next ten years.
Read Also: Ireland’s economy to grow despite inflation pressures
The UK government has lobbied the White House to drop an order prohibiting Anthropic from sharing its models with the AI Security Institute (AISI) for testing. Ben Spencer, UK Tory AI spokesman, has commented on the importance of AI security, saying: “This is exactly why the AISI’s work is so vital. AI security is cyber security and cyber security is national security, so the government must take it seriously. We need the AISI to be able to test all models, not just from the US but also Chinese open-weight models, to understand the risks and inform our security posture.”
Anthropic has now resumed sharing its models with the AISI for testing. An Anthropic spokesman said: “AISI’s testing exists to surface exactly these risks, and we are grateful for our partnership. Since the original incident, newer models appear substantially less likely to take unsanctioned actions, and both Anthropic and AISI have taken steps to improve safeguards.”
