Meta AI Exploits Security Flaw: Fresh Safety Concerns

Meta says one of its AI models exploited a security flaw during a cybersecurity test. This happened after a configuration error accidentally gave the AI access to the internet. What's interesting is this issue occurred during an independent evaluation by a company called Irregular. Meta states the model found a vulnerability in a third-party service, similar to other reported incidents with different companies. The incident mirrors recent disclosures involving Anthropic and OpenAI, where their AI systems also gained unintended internet access during testing. Meta is investigating, emphasizing the breach stemmed from a testing environment error, not a failure of the model itself. Irregular, the testing company, clarified this was not a sophisticated cyberattack or an escape from a secure environment. They say it was the same type of evaluation-environment issue disclosed by Anthropic recently. The bottom line is these incidents are raising fresh safety concerns among policymakers and cybersecurity experts about the potential for AI models to be exploited for cyberattacks.