White House exempts open AI systems

The White House has drawn a line through the U.S. AI market: if a model is “open,” it gets a lighter path; if it is closed and frontier-scale, it faces federal security review. Officials briefed tech companies on Tuesday, August 4, on a framework that would let the government vet cutting-edge systems for cybersecurity risks while excluding open models from that process. (msn.com) That carveout matters because it does not map neatly onto nationality. (nytimes.com) The New York Times reported on August 5 that the administration’s safety plan appears to cover leading U.S. labs such as OpenAI and Anthropic while seeming to exempt some Chinese artificial intelligence models if they are open. (bloomberg.com) The policy lands as money and power in AI remain concentrated. Bloomberg reported on August 5 that Microsoft now generates most of its AI revenue from OpenAI, while Business Insider reported that Google is in talks on a deal worth more than $1.5 billion involving AI coding startup Mechanize. ### Which systems did the White House leave outside review? (msn.com) The administration’s framework, as described by the Washington Post and the New York Times, would apply the voluntary federal review process to closed-source frontier models and exclude models that publish underlying code or key weights. (nytimes.com) Politico reported the same day that the plan would exempt lower-cost, nonproprietary models after a Tuesday White House meeting with major technology companies. Reuters, via U.S. News, separately reported that White House officials discussed the unpublished testing rules with staff from Meta, Anthropic, Google, Nvidia and OpenAI. (bloomberg.com) ### Why are OpenAI and Anthropic at the center of this? The New York Times said the structure of the rules appears to put OpenAI and Anthropic inside the review perimeter because they are major U.S. developers of closed models. That makes the burden company-specific as much as capability-specific. (msn.com) Axios reported on August 5 that the White House’s hands-off approach to open models has shifted attention to how the administration intends to handle Chinese labs. That framing has sharpened questions in Washington about whether the policy’s national-security rationale matches the models it actually captures. ### How does the China issue complicate the policy? (politico.com) Chinese models are part of the dispute because the exemption turns on openness, not origin. The New York Times reported that Chinese AI models appear to be among those effectively spared if they fall into the open category. Axios reported that the White House still has other tools it could use against Chinese open models, even though they sit outside this framework. (nytimes.com) That means the review policy is not the only channel available to Washington, but it is the one now shaping how U.S. and foreign developers are sorted. ### Where do the business stakes show up most clearly? Microsoft’s disclosures offered the clearest number. (axios.com) Bloomberg reported that most of Microsoft’s AI revenue now comes from OpenAI, underscoring how much one partnership still anchors the commercial side of the market. Google’s talks with Mechanize show the other side of the scramble. (nytimes.com) Business Insider reported that Google is discussing a deal worth more than $1.5 billion to license technology and acquire key engineering talent from the AI coding startup, a structure the outlet said can draw less scrutiny than a full acquisition. (axios.com) ### What happens next for companies and investors? August 5 is now the date companies are using to assess the new boundary: open models outside review, closed frontier models inside. The next step is not a single hearing or vote, but how labs, cloud providers and investors respond to the White House framework described this week. (bloomberg.com) Microsoft, OpenAI, Google, Anthropic and companies building open-weight systems will all be measured against that split in the weeks ahead. Investors also have two fresh reference points from August 5 — Microsoft’s disclosure of OpenAI-linked AI revenue and Google’s reported Mechanize talks — as they judge who benefits from the new rules. (nytimes.com) (msn.com) (finance.yahoo.com)