OpenAI and Anthropic Now Need a Federal Sign-Off

Across two days, 26 to 27 June 2026, the US government reshaped how the most capable AI models reach the market. It held OpenAI's new GPT-5.6 family to a small group of approved partners, then one day later let Anthropic put its strongest cybersecurity model back in the hands of critical-infrastructure defenders. Both now pass through the same trusted-partner gate, and that gate rests on an executive order whose formal rulebook does not yet exist.

Two moves, one gate

On 26 June, OpenAI previewed GPT-5.6 at the government's request, available through the API and Codex only to partners "whose participation has been shared with the government." General availability is promised in the coming weeks Projection. Press accounts put the initial group at about 20 organizations, per Axios.

Figures labeled claim are operator-stated and not independently audited; projection marks a forecast or plan, not a built fact.

The next day, Anthropic said the government had notified it that Mythos 5 could return to a set of US organizations that operate and defend critical infrastructure. Fable 5, the public-facing model, stayed dark.

The symmetry is the story. OpenAI moved from an expected open launch to a gated one. Anthropic moved from a total block to a partial restore. Both landed on the same arrangement, with access decided, in effect, in Washington.

The executive order the coverage flattened

The legal backdrop is Executive Order 14409, signed 2 June 2026. Section 3 instructs agencies, within 60 days, to roughly 1 August Projection, to build a classified process that defines when a model becomes a "covered frontier model," with the determination made by the Director of the NSA.

The same section directs agencies to design a voluntary framework under which a developer could give the government access to a covered model for up to 30 days before releasing it to other trusted partners. Much of the week's coverage compressed this into a 30-day government review before public release, which is not what the text says.

Section 3(c) is explicit: nothing in it authorizes "a mandatory governmental licensing, preclearance, or permitting requirement" for releasing AI models. The current per-customer sign-off for GPT-5.6 is therefore an ad hoc measure filling the gap before the framework exists, not an exercise of it.

Mythos 5 returns, Fable 5 stays dark

In its own statement, Anthropic said the government had told it that Mythos 5, its strongest cybersecurity model, could be redeployed to a set of US organizations that operate and defend critical infrastructure, and that it would continue working to expand Mythos 5 access and to make Fable 5 generally available again.

"We don't believe this kind of government access process should become the long-term default."

OpenAI, 26 June 2026

On the government side, reporting describes a letter from Commerce Secretary Howard Lutnick to co-founder and Chief Compute Officer Tom Brown, dated 26 June and seen by Bloomberg, determining that "appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model," and noting that Anthropic's work had yielded significant progress.

Press accounts (Semafor and Reuters, via TechCrunch) put the cleared group at more than 100 US agencies and companies, and say non-American staff at those organizations, including Anthropic's own, may now use the model. Before the 12 June block, roughly 200 firms had Mythos access through Anthropic's Project Glasswing. The Lutnick letter reportedly does not mention Fable 5, and talks were said to continue toward restoring it Projection.

What GPT-5.6 is, and why Washington flinched

GPT-5.6 arrived as three tiers under a new naming scheme, where the number marks the generation and the names mark durable capability tiers. List prices, per 1 million tokens:

The concern is cyber capability. OpenAI calls Sol its strongest cybersecurity model yet but frames it as a defender, saying it is "better at helping people find and fix vulnerabilities than reliably carrying out end-to-end attacks." In Chromium and Firefox tests it found bugs and exploitation primitives but did not autonomously build a full-chain exploit, and OpenAI says it stays below the "Cyber Critical" threshold in its Preparedness Framework.

OpenAI's own benchmarks Claim place Sol at the front of agentic coding. On Terminal-Bench 2.1 it reports Sol Ultra at 91.9 percent and standard Sol at 88.8 percent, against 88.0 percent for Claude Mythos 5 and 70.7 percent for Google Gemini 3.1 Pro Preview. On its ExploitBench, OpenAI says Sol is competitive with Anthropic's Mythos Preview while using roughly a third of the output tokens. These figures are vendor-reported and not independently audited.

Why it matters

For the first time, the release of leading US models is being decided case by case in Washington, not by the labs alone. Both companies have objected. OpenAI called the process unsustainable and warned it keeps tools from developers, enterprises, and global partners. Anthropic, in the post announcing its shutdown, said a blanket version of the standard would "essentially halt all new model deployments for all frontier model providers."

The unresolved tension is procedural. A de facto approval regime is running today, while the order meant to govern it both disclaims mandatory preclearance and remains unbuilt until about 1 August Projection. Until the framework lands, each release is negotiated in the gap.

Sources

1. OpenAI, "Previewing GPT-5.6 Sol: a next-generation model", 26 June 2026.
2. OpenAI Help Center, "A preview of GPT-5.6 Sol, Terra, and Luna", 26 June 2026.
3. The White House, "Promoting Advanced Artificial Intelligence Innovation and Security" (Executive Order 14409), 2 June 2026.
4. Anthropic, statement on Mythos 5 redeployment (via Fortune and TechCrunch), 27 June 2026.
5. Fortune, "Anthropic's Mythos 5 AI model cleared by U.S. for wider use", 27 June 2026.
6. TechCrunch, "Trump Admin releases Anthropic Mythos to be used by more than 100 US companies, agencies", 26 June 2026.
7. The Next Web, "US clears Anthropic to restore Mythos 5 access for trusted cyber defenders", 27 June 2026.
8. Axios, "OpenAI releases powerful new GPT-5.6 model under restrictions", 26 June 2026.
9. TechCrunch, "OpenAI limits GPT-5.6 rollout after government request", 26 June 2026.
10. CNN Business, "White House asks OpenAI to limit its next model release", 25 June 2026.