OpenAI delays GPT-5.6 release after a US administration request
OpenAI is staggering the release of its new AI model, GPT-5.6. According to the Verge, the decision came after the US administration requested that the model be shared in stages before being made available to the broader public, citing potential security concerns.
According to the reports, OpenAI plans to share the model first with only a select group of partners, leaving access for the wider user base to a later stage. This approach foresees a controlled rollout of a new model rather than opening it directly to everyone.
Such staggered releases are not new in the AI sector. Developers often test large models in real-world conditions by first opening them to a limited group of users. This process aims to detect a model's unexpected behaviours and possible vulnerabilities before they spread widely.
According to the Verge, what is different this time is the rationale behind the delay. The report states that the US administration, cautious about potential security issues, asked for the model's release to be staggered. This stands out as an example of a government taking a direct interest in the release schedule of advanced AI models.
The safety of AI models has become both a technical and a political subject of debate in recent years. The potential for misuse of the most powerful models is assessed across a wide range, from cybersecurity to the generation of misinformation. How and when models are released is therefore being treated with ever greater care.
Governments becoming involved in this process is a significant development for AI governance. Until now, model release decisions have largely rested on the internal assessments of the developer companies. An administration making a request about release timing suggests that public authorities may be seeking a more active role in this field.
That said, the form of such arrangements is not yet clear. Whether it is a formal rule or a voluntary cooperation remains ambiguous in the details of the report. This uncertainty shows that the relationship between government and companies in the AI field is still taking shape.
For OpenAI, the delay could have both technical and commercial consequences. Releasing a model later can matter for timing in the competition with rivals; but a cautious approach to safety can also be viewed in terms of building trust over the long run.
For users, the practical effect will be a delay in access to the new model's features. A staggered release means that only certain partners benefit from the new capabilities at first, while the broader public has to wait.
The development reported by the Verge is a sign that AI has become not just a technology product but also a matter of public policy. Model release decisions becoming intertwined with security assessments offers clues about how the sector may take shape in the period ahead.
Read next
How can video games train AI agents? General Intuition's $2.3B bet
An AI startup has raised $320 million to train AI agents using millions of hours of gameplay data, reaching a $2.3 billion valuation. The idea is that action data from games can give AI a sense of intuition useful in the real world.
A planet orbiting so close its magnetic field connects to its star: how does it work?
Astronomers have found a planet orbiting so close to its star that the two magnetic fields connect. At the right point of the orbit and stellar cycle, the star's atmosphere brightens. Here is the science of this unusual system.
Why is AI's power bill so high, and can it be cut 1,000x?
Databricks' former AI chief is working on a new approach that he says could cut AI's energy consumption by up to 1,000 times. The claim revives the question of why AI uses so much energy and how that could be reduced.
What is IBM's sub-1 nanometer chip technology, and why does it matter?
IBM says it has developed the world's first sub-1 nanometer chip technology. The structure, called nanostack transistors, could boost chip performance or energy efficiency. Here is what the claim means and why it matters.
How liquid cooling cuts data-center water use to near zero: a simple explainer
A new data-center cooling design that runs warm, at around 45 degrees Celsius, can cut water use to almost nothing while keeping AI chips cool. This explainer breaks down why data centers consume so much water, how warm liquid cooling changes the equation, and what it means as AI demand grows.