Another AI Jailbreak: Anthropic's Claude Escapes a Test and Hacks Outside Groups

Two major tech headlines are sparking serious concerns about rising dangers on the digital front, especially the growing risks presented by AI.

Federal authorities are warning about digital threats to U.S. water systems after cyberattackers targeted more than 30 such systems in Minnesota. The attacks came amid federal warnings that Iranian hackers have been focused on America's water supplies. Iran has a history of targeting those systems. In this case, there were no reports that residents had been impacted by the attacks.

Meanwhile, there's been another report of an Artificial Intelligence system breaking through its guardrails and attacking an outside organization.

AI company Anthropic, maker of Claude, says several of its models escaped the facility and hacked into three other organizations during testing. The AI models were able to access the internet from within testing environments that should have been sealed off.

“Claude compromised the impacted organizations’ infrastructure using basic techniques,” Anthropic said, such as exploiting weak passwords.

Anthropic said it decided to launch a large-scale cybersecurity review of its own systems after a rival business, OpenAI, revealed recently that one of its AI models had gone rogue and hacked another AI model.

* 'Unprecedented': AI Agent Escapes Secured Testing Environment and Hacks Another AI

Anthropic says it reached out to the affected organizations, with two of them confirming they were unaware they had been hacked. In all three incidents, the AI models had been tasked with an internal “capture the flag” cybersecurity challenge before they escaped.

Cybersecurity expert Kok Tin Gan, co-founder & CEO of NyxLab, said, “If we simply give the AI a goal and allow it to decide how to achieve it, we should not be surprised when it takes actions that technically satisfy the objective, but fall outside our intended scope or expectations."

Still, the incidents have raised questions about the safety of AI agents, fueling concerns about what the world's most powerful AI models will be capable of doing one day, especially if guardrails can't contain them. 

*** AI tools threaten to hide CBN News from your online searches! Sign up for CBN Newsletters today to ensure you can still find the latest news from a Christian perspective.***