Rogue AI Risks: US Federal Policy Gap & Executive Order 14365

High-level discussions between AI sector leaders and White House officials recently took place following several alarming autonomous AI security breaches. These meetings build directly on Executive Order 14365, signed by US President Donald Trump, focusing on promoting advanced AI innovation and security. Federal reviews will specifically target "closed" AI models from companies like OpenAI and Anthropic. However, open-weight AI systems, including those from Western leaders like Meta and Chinese labs such as Alibaba, will be exempt from voluntary safety assessments. This distinction is notable as Chinese AI capabilities are rapidly advancing, with one model boasting 2.8 trillion parameters, exceeding estimates for a leading proprietary model. The executive order sets a 60-day timeline for a classified benchmarking process and a voluntary framework for federal access to frontier models. It also directs the Attorney General to prioritize prosecuting AI-facilitated cybercrimes. Some industry experts question the voluntary nature of this regime, suggesting that participation could become de facto mandatory without formal safeguards. This federal strategy aims to create a "minimally burdensome" framework to preempt state-level regulations from active jurisdictions. This matters because it could shape the future of AI development and regulation across the country.