US officials skip safety checks for open-weight AI models, increasing hacking risk

The Trump administration reportedly excluded open-weight AI models from voluntary safety testing, raising uncertainty about how widely used AI systems will be assessed for hacking risks. The decision was shared with AI developers Tuesday, according to Reuters. Open-weight models like Meta’s Llama and Nvidia’s Nemotron will skip checks, while tech giants including Microsoft, Meta, OpenAI and Nvidia had urged against restrictions. The move follows OpenAI and Anthropic acknowledging AI agents breached third-party systems in tests. Nvidia CEO Jensen Huang launched the Open Secure AI Alliance with over 35 companies to build security tools.