OpenAI’s GPT-5.6 cleared for launch after U.S. security review delayed release

OpenAI will be releasing its latest version, GPT-5.6, this Thursday, after the U.S. government compelled it to postpone the release earlier this month due to security concerns. According to a source, the U.S. Department of Commerce has approved the launch of this product, though initially only government-sanctioned organizations had access to it.

The green light comes closely on the heels of the reinstatement of Anthropic’s Fable and Mythos models, which the government had also paused for nearly three weeks over similar safety concerns.

Authorities were particularly concerned about the models’ ability to uncover software vulnerabilities that cybercriminals could exploit.

OpenAI will launch all three variations of GPT-5.6

The launch had been postponed last month after the U.S. government requested that OpenAI temporarily limit access to GPT-5.6 to a small group of vetted partners.

Officials were concerned that frontier AI models may be exploited malicious purposes if released without additional safeguards.

The approval comes as Washington develops a new framework for overseeing frontier AI systems. In recent weeks, the Trump administration has worked with leading AI developers to establish voluntary safety standards. The U.S. government gave GPT-5.6 the green light following a round of rigorous state testing dictated by the newly established oversight framework for advanced AI.

The GPT-5.6 comes in three versions tailored for various needs. For one, there is the high-grade Sol, the mid-tier Terra, and the fast and cheap Luna. On X, the company announced that all three models would be publicly accessible on Thursday. In addition, Terra would cost only half as much as GPT-5.5, as it works to win more clients.

When the pause was imposed, OpenAI assured users it would only be temporary and that keeping the technology locked away would be counterproductive. It stated, “We don’t believe this kind of government access process should become the long-term default.”

At the time, it also stated that it would work hand in hand with government bodies to develop the cyber Executive Order framework and to design a repeatable blueprint for next-generation releases.

Since President Donald Trump assumed office, the White House has said it wanted to prioritize rapid innovation over strict oversight, believing that minimal red tape would help the U.S. defeat China in the AI race. Still, it has raised concerns that foreign adversaries could misuse the new model releases, increasing scrutiny.

At the moment, the federal government is defining the specific technical thresholds that will subject frontier AI systems to new security restrictions under the recent executive decree.

Anthropic promises to find vulnerabilities proactively and report any hacks

Last week, Anthropic brought its Fable 5 and Mythos 5 models back online after Washington retracted its restrictions, but Mythos 5 is available only to a select group of trusted domestic partners.

Aside from the model’s greenlight, U.S. Commerce Secretary Howard Lutnick contended that Anthropic had committed to proactive risk management, continuous collaboration on safety benchmarks, and the immediate reporting of any weaponized use.

Secretary Lutnick, however, said that the Commerce Department had the power to revoke this clearance and reinstate licensing mandates if circumstances change or Anthropic fails to meet its obligations.

According to a source, the firm has already added a new safety feature to stop a hacking technique that worried the US government, even though the dev team had already patched the actual vulnerabilities weeks ago.

So far, security experts claim that the Mythos models could make it much easier for malicious actors to launch sophisticated cyberattacks, especially against financial institutions with legacy systems.

However, some critics have pushed back against the administration’s handpicking of which companies qualified as “trusted.” Some claimed the process was unfair and not transparent.

Meanwhile, the financial stakes for AI dominance have soared, and OpenAI and Anthropic have now submitted their IPO documents targeting the $1 trillion mark.

If you're reading this, you’re already ahead. Stay there with our newsletter.