Google, Microsoft, and xAI Agree to Share Early AI Models with US Government

In a new voluntary agreement reported by the Wall Street Journal, Google, Microsoft, and xAI (Elon Musk's AI company, creator of Grok) have committed to sharing early versions of their AI models with the US government. The move is part of a broader effort to allow federal safety evaluation before public deployment.
Key Details
- The agreement involves providing the US government with early access to AI models for pre-release safety testing and risk assessment.
- Companies are acting voluntarily, not under a legal mandate. The goal is to establish a framework for responsible AI development and catch potential harms early.
- Specific models affected were not named in the WSJ article, but it likely covers major upcoming releases from each company: Google's Gemini upgrades, Microsoft's Copilot/OpenAI integrations, and xAI's Grok iterations.
- This is reminiscent of earlier voluntary commitments made by AI labs to the White House in 2023, but now formalized into a more structured sharing agreement with government agencies.
For developers using AI coding agents (e.g., GitHub Copilot, Claude, or local models like Grok), this means the models you deploy may have undergone federal safety review. It could also signal more regulatory oversight ahead, potentially affecting API access and model behavior guardrails. Keep an eye on how these evaluations impact model release timelines and capability restrictions.
The HN discussion (36 points, 57 comments) on this article is active here.
📖 Read the full source: HN AI Agents
👀 See Also

OpenClaw's New Release: A Simple Name Change or a Major Upgrade?
OpenClaw, previously known as ClawDBot, has undergone a transformation. Read on to find out whether this change is merely cosmetic or introduces new features and improved stability.

Minimax M2.7 and Scaling to 100k+ OpenClaw Instances Discussed in Ecosystem Session
Jim and AndyML hosted the Minimax team to discuss Minimax M2.7 and how they scaled their hosting environment to support over 100,000 OpenClaw instances. The session attracted 100-110 users from Discord and 350,000+ viewers on a Chinese simulcast.

AlphaEvolve: DeepMind's Gemini-powered agent optimizes algorithms across genomics, power grids, and TPC circuits
AlphaEvolve, a Gemini-powered coding agent by Google DeepMind, improved DeepConsensus variant detection errors by 30%, boosted AC Optimal Power Flow GNN feasibility from 14% to 88%, and reduced quantum circuit error by 10x.

‘AI Washing’: UK Firms Rebrand as AI Companies Despite Weak Links
PR executives report UK companies forcing them to pitch ordinary automation as AI, with 50% of AI-related press releases sent under duress. Examples include an AllBirds pivot to acquiring AI GPUs and a property firm calling a handheld scanner an AI tool.