White House now vets who gets OpenAI, Anthropic frontier models
TL;DR
- Lutnick personally approved roughly 20 OpenAI customers during GPT-5.6's limited preview, establishing individual-level gatekeeping with no published criteria.
- OpenAI publicly stated the access process should not become the long-term default, signaling labs view current compliance as coerced rather than voluntary.
- The Anthropic episode set the first precedent for frontier model suspension via export-control directive, covering even the company's own foreign-national employees.
A CNBC scoop this week reports that the Trump administration has started deciding, or at least strongly signalling, which companies get to touch the newest frontier models from OpenAI and Anthropic. The mechanism reportedly runs through Gold Eagle, a cyber-vulnerability clearinghouse the White House recently launched, and it would layer government sign-off on top of the partner lists both labs already curate for their most capable systems.
The specific vehicles are worth naming. Anthropic's Project Glasswing gives a handful of partners early access to its Mythos cybersecurity model. OpenAI runs a similar consortium called Daybreak. CNBC's reporting is that going forward, adding a new partner to either roster requires explicit government approval, even as a White House official told the outlet that participation is voluntary and that decisions on timing and scope of releases rest entirely with the companies.
That 'voluntary' framing is doing a lot of work in a month that included a very involuntary-looking episode. In mid-June, Anthropic disabled Claude Mythos 5 and Fable 5 to comply with an export-control directive citing national-security authorities, and access only came back after weeks of negotiation, with Commerce Secretary Howard Lutnick clearing Mythos 5 for a select group of companies and federal agencies. OpenAI, per CNBC, was separately asked to gate its recent GPT-5.6 release. Take those together and the direction of travel is clear enough even if any single case is technically opt-in.
If you buy or build on frontier models, the practical shift is that your vendor is no longer the only party in the room deciding whether you get the next tier of capability. Procurement timelines have a new dependency, and being inside an approved consortium starts to look like a strategic asset in its own right. The honest caveat is that the reporting leans on unnamed sources, the White House disputes parts of the framing, and none of the coverage spells out the criteria the administration is actually applying or how long the labs will tolerate the arrangement.
The upside case, if you squint, is for buyers already inside Glasswing or Daybreak and for providers of open-weight alternatives, since anyone spooked by Washington-shaped release calendars now has a reason to hedge.
What others are reporting
-
SecurityWeek Read →
Adds Stanford expert pushback on the policy's factual basis and documents the Pentagon's separate designation of Anthropic as a national security risk, revealing contradictory government posture.
"We don't believe this kind of government access process should become the long-term default," OpenAI said in a statement.
-
The Next Web Read →
Frames Gold Eagle as de facto mandatory despite its voluntary branding and adds geopolitical framing on Chinese competitors closing the capability gap as US distribution tightens.
If labs cannot release capable models without government approval of the partner list, the US government has acquired de facto distribution authority over frontier AI.
-
Benzinga Read →
Juxtaposes US gatekeeping with Kimi K3's simultaneous release and Sacks's race-to-the-bottom warning, framing customer access control as a competitive vulnerability with market implications.
"The Administration continues to collaborate with all of America's frontier labs to strengthen the security of this technology."
-
Spencer Fane Read →
Five concrete contract-drafting recommendations for enterprise AI customers: force majeure for regulatory suspension, vendor diversification requirements, and explicit export control representations.
"Commerce reserves the right to reevaluate the decisions made in this letter...should Anthropic fail to adhere to its commitments."
Originally reported by cnbc.com
Read the original article →Original headline: White House Now Dictating Which Customers Get Access to Frontier AI Models From OpenAI and Anthropic