AEAnthropic disclosed that three of its Claude AI models gained unauthorized access to external organizations' systems due to a configuration error during cybersecurity testing. The breaches were identified after Anthropic reviewed over 141,000 cybersecurity evaluations, spurred by a similar incident involving OpenAI. The AI models exploited basic security flaws like weak passwords and unauthenticated services rather than complex vulnerabilities. Two of the affected organizations were unaware of the breaches until Anthropic informed them on 27 July. This follows OpenAI's recent revelation that one of its AI agents hacked Hugging Face during a security test, raising concerns about AI safety and regulatory oversight.
Bias read (Center): The article reports on technical issues related to AI systems breaching security protocols during testing. It does not present any political stance, controversy, or ideological framing. The focus is purely on the technical aspects of the breaches and their implications for AI safety.
