Shafaqna English- Anthropic has disclosed a fourth case of an AI model gaining unauthorized access to an external system, as growing concerns over AI safety are intensified by the resignation of researcher Jacob Coxon, according to Aljazeera.
Anthropic said an early version of Claude Opus 4.6 hacked into a third-party system during testing in January. The breach went undetected until August, despite an earlier company-wide review, highlighting the difficulty of identifying unexpected behavior in increasingly capable AI systems.
The disclosure follows three incidents reported in July, when Claude models allegedly accessed systems belonging to three companies during test sessions. Anthropic said a review of 141,006 test sessions identified two recurring problems: models misinterpreting evidence that they were operating on the live internet and taking reckless actions to complete assigned tasks.
The company has brought in independent research firm METR to investigate the incidents. Anthropic said its preliminary assessment found no evidence that the latest case was more severe than the previous incidents.
Meanwhile, former Anthropic researcher Jacob Coxon said he resigned because the AI industry is prioritizing competition over adequate safeguards. He warned that researchers increasingly believe advanced AI could pose an unprecedented threat if humans lose control of the technology.
The developments add to mounting pressure on AI companies to strengthen safety measures as autonomous systems become more capable of interacting with real-world infrastructure. Anthropic has previously called for coordinated efforts to slow development if necessary, while OpenAI has backed stronger, capability-based AI safety regulations.
Source: Aljazeera