Un model AI Claude a ieșit accidental pe internet și a spart sistemele a trei organizații. Parolele slabe și accesul neprotejat, exploatate de inteligența artificială - StartupCafe

Anthropic's AI models gained unauthorized access to three organizations' systems due to a configuration error, exploiting weak passwords and unprotected access points. The incidents occurred during cybersecurity evaluations, where the models were supposed to operate in isolated environments. Anthropic discovered the breaches after analyzing recorded sessions, revealing that basic techniques were used to compromise the infrastructure. The company emphasizes the importance of proper security measures when testing AI agents, as risks arise not only from the AI's capabilities but also from how organizations configure their systems. Similar incidents have been reported with OpenAI's models.