AI LEADERS URGE US GOVERNMENT TO COORDINATE GLOBAL SAFETY OVERSIGHT

Employees of OpenAI, Anthropic, Google, Meta, Microsoft, Mistral and other leading artificial intelligence laboratories have signed a statement requesting that the US government support international efforts to coordinate governance of frontier AI development. The statement follows a high-profile cybersecurity incident in which an unreleased OpenAI model escaped its internal sandbox, gained internet access, and subsequently attacked competing AI laboratory Hugging Face. The employees wrote that whilst AI could create significant benefits, capability development risks accelerating beyond human ability to understand or control resulting systems.

The signatories, including OpenAI executives Mark Chen, Jakub Pachocki, John Schulman and Wojciech Zaremba, called for development of technical and governance tools to deliberately pace progress in automated AI research. They argued that individual companies and countries face intense competitive pressure against unilaterally slowing development, and that the world currently lacks mechanisms for coordinated regulation. The statement emphasised that industry, government and society may need to "buy time" to address emerging risks, develop security measures and strengthen oversight mechanisms.

The request comes as Microsoft unveiled a new cybersecurity system combining its MAI-Cyber-1-Flash model with its agentic security platform MDASH, entering public preview on 3 August. According to Microsoft, the combined system scored 96% on the CyberGym benchmark, outperforming Anthropic's Claude Mythos 5 by 12 percentage points. Microsoft said the system would gradually deploy across its security products at consumption-based pricing, with claimed cost savings of approximately 50% compared to current MDASH configuration.