On a single Friday, the Trump administration gated OpenAI’s launch of GPT-5.6 Sol, Terra, and Luna to roughly 20 government-approved companies and, in the same news cycle, cleared Anthropic to restore its Mythos 5 cybersecurity model to about 100 federal agencies and critical-infrastructure operators. Two of the country’s strongest frontier labs are now shipping product through customer lists screened in Washington.

The instrument is a letter. Commerce Secretary Howard Lutnick wrote to Anthropic clearing Mythos 5 for “certain trusted partners,” per correspondence seen by Bloomberg. Restrictions on Fable 5 weren’t lifted. Weekend talks continued. The underlying authority traces to a Trump executive order from early June requiring safety testing of advanced models, a mechanism NBC News reports “remains in development.”

Both companies are publicly hedging. OpenAI said in a blog post that “We don’t believe this kind of government access process should become the long-term default,” even as it works with the administration on what it calls a “repeatable process” for future releases. Sam Altman called the staggered debut “bad news” on X. At Anthropic, co-founder Tom Brown handled the negotiations with Lutnick directly; CEO Dario Amodei, Bloomberg reports, took a “hands-off role.”

The historical rhyme is the late-1990s Wassenaar-era cryptography export controls, where the State Department effectively decided which customers could receive strong crypto until the regime collapsed under its own ad-hocery in 2000. The current arrangement has the same architecture: a pre-clearance process built on phone calls and letters, applied to the most capable models, without a rule on the books.

What’s revealing isn’t the safety logic. It’s how quickly both labs accepted being intermediated. The “long-term default” line reads less as resistance than as a marker laid down for the record.

Sources

Sources