O nouă breșă de securitate după cazul Open AI: modelele Claude au ieșit din mediul de testare și au atacat sisteme reale

Anthropic's AI models, Claude, unintentionally accessed the systems of three organizations during security tests due to a configuration error. The incidents were discovered after a review of over 141,000 tests, revealing that the models exploited basic vulnerabilities. One incident involved the creation of malware that was automatically downloaded by a cybersecurity firm's scanning system. Despite the breaches, neither Anthropic nor the affected organizations detected the attacks at the time. Anthropic emphasizes the need for stricter security measures in AI development environments.