OpenAI releases ChatGPT 5.6 after White House delay

ChatGPT 5.6 reached the public on Thursday after OpenAI says it spent weeks limiting access at the White House’s request. According to The Guardian, the Trump administration asked last month that the model be released only to a small group of government-approved users, and OpenAI complied while officials were briefed on its capabilities.

The episode shows how the US government is turning distribution into a control surface for advanced AI. A model release is no longer just a product decision; it can be treated as a security-sensitive event, with access granted to “trusted partners” first and everyone else later. Axios cited by The Guardian describes additional testing by a government body called the Center for AI Standards and Innovation before the wider rollout. This staggered pattern is starting to look like a template: the Guardian links it to similar restrictions placed on Anthropic’s latest models last month, including a temporary export ban.

For OpenAI, the short-term cost is commercial. Limiting a headline model to approved users slows adoption, delays feedback from the broader market, and risks pushing customers toward rivals that can ship faster. But the alternative is to collide with a government that can threaten export controls, procurement lockouts, or informal pressure on partners. The Guardian reports that the administration has otherwise encouraged rapid AI development, framing it as an arms race with China, while intervening when fears of AI-driven cyberattacks become harder to ignore. That combination—public rhetoric about speed, private demands for restraint—leaves companies guessing what will trigger the next call from Washington.

The politics also travel. When the US restricts access to the most advanced models, allies and industries that rely on them are forced into the same queue, regardless of their own regulatory choices. The Guardian notes tensions with US allies over sharing AI breakthroughs, and describes concern in sectors that depend on these tools for business. In practice, the country that hosts the leading labs can decide which foreign firms get frontier capabilities first, and which must wait until the risk has been “tested” by agencies that answer to domestic priorities.

OpenAI says ChatGPT 5.6 is its safest and most capable model yet, and the Guardian reports it includes a flagship product called Sol. The model still launched only after a period in which access was rationed by government request, then expanded once officials were satisfied.

On Thursday, the new model arrived as a public release. The gatekeeping happened earlier, in the login list.