AssetID: 55890400
Headline: UNCAPTIONED: Anthropic Says Claude AI Breached Three Companies During Testing
Caption: Anthropic Says Claude AI Breached Three Companies During Testing. Anthropic says its Claude AI accessed and hacked three companies during cybersecurity tests after a configuration error. The mistake accidentally gave Claude internet access, despite testing environments being designed to remain completely isolated from outside networks. Anthropic launched an internal review after OpenAI recently disclosed a separate rogue AI security testing incident. Investigators examined more than 141,000 test sessions before identifying three unauthorized breaches involving separate Claude models. The company said Claude exploited weak passwords and unsecured systems using relatively basic hacking techniques during evaluations. Anthropic suspended all cybersecurity evaluations immediately, identified the affected organizations, and began notifying them about the unauthorized activity. The company said the incidents highlight the need for stronger safeguards as AI systems gain increasingly advanced cyber capabilities. Instructions: THIS VIDEO MUST NOT BE EDITED FOR LENGTH TO COMBINE WITH OTHER CONTENT
Keywords: Current Affairs and Politics,Anthropic,Claude AI,artificial intelligence,AI cybersecurity,cybersecurity,AI hacking,cyberattack,internet access,OpenAI,rogue AI,security testing,weak passwords,cybersecurity testing,AI safety,cyber capabilities,technology
PersonInImage: