White House Nears Pre-Release AI Review Deal with Five Major Labs

The White House is close to finalizing voluntary safety standards with Anthropic, OpenAI, Google, Microsoft, and Amazon, including a 30-day government review for frontier models and a new jailbreak severity scoring scale. Separately, Abu Dhabi's MGX closed a record $49 billion fund backing all major US frontier labs, Anthropic launched its own drug discovery programs and poached a Nobel laureate from DeepMind, Microsoft deployed $2.5 billion to embed 6,000 engineers inside enterprise clients, and Stripe enabled AI agents to spend real money autonomously.

Written by AI from the sources it cites; it can be wrong — follow any citation to confirm.

7 sources1 min read

White House and five labs target August 1 for AI safety deal with jailbreak scoring scale The framework would open a 30-day government review window for frontier models, with NSA/CISA access before release. A new shared jailbreak severity rubric, modeled on CVSS, would give the industry a common language for rating vulnerabilities. 1

Abu Dhabi's MGX closes $49B fund, backing every major US frontier AI lab The hybrid sovereign-institutional vehicle co-led rounds at Anthropic, OpenAI, and xAI, and now targets $100 billion in assets. The fund positions Gulf capital as the indispensable external financier of frontier AI development. 2

Anthropic launches drug discovery programs and poaches Nobel laureate from Google DeepMind Anthropic announced Claude Science for researchers and its own drug development for neglected diseases. The company also hired DeepMind director and 2024 Nobel laureate John Jumper, plus two key Gemini contributors, ahead of its anticipated IPO. 3 4 5

Microsoft launches $2.5B 'Frontier Company' with 6,000 engineers to fix AI pilot failures The forward-deployed engineering unit will embed Microsoft staff inside enterprise customers to design, build, and operate AI systems on-site, targeting the 95% of enterprise AI pilots that deliver zero measurable profit impact. 6

Stripe and Cross River enable AI agents to spend real money autonomously The partnership provides bank-grade single-use card issuance for AI agents, allowing autonomous software to make payments without exposing user credentials. This marks a key step in building payment infrastructure for agentic commerce. 7