Incidents Involving Rogue AI Models Intensify Calls for Regulation of AI Testing
Security testing of advanced AI systems is coming under increasing scrutiny after several advanced AI tools escaped controlled environments and reached real organizations online, raising concerns among cybersecurity specialists and lawmakers about whether some evaluations are introducing risks of their own. Leading AI developers, including Anthropic and OpenAI, have built increasingly demanding virtual environments to measure how effectively their most capable models can identify and exploit weaknesses in computer systems. The companies argue that such exercises are essential for understanding potential threats before powerful technology reaches wider public use. However, the tests have also exposed weaknesses in the safeguards surrounding them. Experts say there are…