OpenAI on Thursday released GPT-5.6, a new generation of the models behind ChatGPT and its coding tools, together with a product for businesses called ChatGPT Work. The company pitched the models as more capable at software development and other complex tasks, The Verge reported. But the most notable feature of the launch was not a technical one: it came after a review by a US government body.
What launched
GPT-5.6 arrives as a family of models pitched at different needs and price points, from a top-end version aimed at demanding, multi-step "agentic" work to cheaper, faster options for everyday use, TechCrunch reported. OpenAI highlighted gains in coding and said the models were its strongest yet on cybersecurity tasks such as finding and fixing software flaws. It also introduced ChatGPT Work, an agent aimed at businesses that is designed to produce finished documents, spreadsheets and simple apps and to carry out longer tasks with less hand-holding, rolling out first to higher-tier subscribers. The specific performance figures cited are OpenAI's own.
The government review
What set this release apart was the process before it. OpenAI said a US government body, part of the Commerce Department that focuses on AI standards, spent about two weeks testing the models before they went out, with the company's engineers working alongside officials. OpenAI stressed that the review was voluntary in legal terms but said it functioned, in practice, like a pre-clearance, an unusual step for a commercial software release. The central concern, by OpenAI's account, was the models' growing ability in cybersecurity, capabilities that can help defenders but could also aid attackers.
Why it matters
That framing is significant. Governments have debated for years how, and whether, to vet powerful AI systems before release, and formal rules remain patchy. A voluntary review that behaves like an approval sits in a gray zone: it gives officials a look at frontier models without a binding legal regime behind it. OpenAI itself said it did not think such reviews should become the permanent default, arguing they could slow useful tools from reaching users and defenders. Critics of the industry, by contrast, have long called for more independent scrutiny, not less, of exactly these capabilities. Newsparlor could not independently verify the technical claims OpenAI made about the models.
The competitive backdrop
The launch also sharpens a business contest. With ChatGPT Work, OpenAI is pushing further into the workplace, competing with rivals, including Anthropic, that are building AI "agents" meant to do office tasks rather than just answer questions. The wider bet across the industry is that these systems will move from chat toward doing, handling real work across a person's tools and files. GPT-5.6, and the way it was cleared for release, is a marker of both that ambition and the unresolved question hanging over it: who, if anyone, checks these systems before they reach the world, and on what authority.
