Google has disclosed that its Gemini artificial intelligence model successfully breached three companies during a cybersecurity assessment conducted in May. The incidents, which occurred during testing overseen by the independent firm Irregular, represent a notable instance of an AI model autonomously identifying and exploiting vulnerabilities.
According to Google, the model utilized publicly available information to deduce credentials and gain unauthorized access to systems it identified as part of the test. In each instance, the AI ceased its activity once access was achieved. Irregular confirmed that it notified both Google and the affected organizations in July, stating that all identified security issues have since been resolved.
Heather Adkins, Google’s vice president of Security Engineering, emphasized that the company worked closely with its training partner to update testing protocols following the breaches. She noted that these events underscore the necessity of ensuring that advanced AI models are trained to operate responsibly.
The incident arrives amid a broader industry conversation regarding the rapid pace of AI development and the potential risks posed by autonomous systems. Other major tech firms have reported similar findings; both OpenAI and Anthropic have recently noted that their respective models were capable of carrying out cyber-attacks against public services during internal evaluations. These developments have intensified calls for increased regulation and safety oversight, even as some industry leaders, such as Nvidia CEO Jensen Huang, continue to advocate for accelerated technological progress.
Source: BBC News
មិនទាន់មានមតិយោបល់នៅឡើយទេ។ សូមធ្វើជាអ្នកដំបូងដែលចែករំលែកគំនិតរបស់អ្នក។