GPT-5.6: When Government Steps In to Control the Release of the Most Powerful AI Models

Government Restrictions on the Latest AI Models

In a striking move that reflects the growing tensions between the AI industry and the U.S. administration, OpenAI announced that it has restricted access to its new GPT-5.6 model family. The models are now available only to a limited group of trusted partners whose participation has been disclosed to the government — a step taken in response to a formal request from the U.S. administration.

What Does the New GPT-5.6 Family Include?

The GPT-5.6 family consists of three distinct models, each designed for different use cases and price points:

• Sol: The flagship and most capable model, equipped with advanced agentic capabilities in coding, biological research, and cybersecurity. It supports multiple reasoning modes, including an "ultra" level that deploys parallel agents to tackle highly complex tasks.
• Terra: A balanced model suited for everyday use, offering solid performance at a reasonable cost.
• Luna: The fastest and most affordable option, ideal for applications that require real-time responses and high-volume generation.

On pricing, Sol starts at $5 per million input tokens and $30 per million output tokens, while Terra offers half those rates. Luna comes in at just $1 per million input tokens and $6 per million output tokens. The company has also introduced notable improvements to prompt caching to reduce recurring costs.

OpenAI's Stance on Government Intervention

Despite complying with the government's request, OpenAI has been candid about its reservations. In an official post, the company stated that this approach denies users, developers, enterprises, and international partners access to tools they genuinely need. It stressed that the current situation must remain a temporary exception, not a permanent norm. OpenAI is seeking to work with the U.S. administration to establish a clear cybersecurity regulatory framework that defines how future models are released — in a transparent and repeatable manner.

Deeper Concerns: Governance vs. Innovation

This situation raises a fundamental debate about the limits of government authority in regulating the pace of AI development. Dean Ball, a former White House advisor who recently joined OpenAI, revealed that the latest executive order — which requires AI companies to brief the government on their models 30 days before release — has effectively created a de facto mandatory licensing system, even if it is nominally labeled "voluntary."

Ball argues that the absence of objective, transparent safety standards could open the door to repeated delays in model releases, potentially undermining the United States' competitive edge in the AI race with China and threatening the returns on the massive investments already committed to technical infrastructure.

The ramifications extend beyond OpenAI alone. Anthropic was forced to fully withdraw its most capable model after the U.S. administration demanded it revoke foreign nationals' access — a requirement that proved harder to implement than pulling the model entirely. A subsequent technical audit revealed that the model had been covertly rerouting high-sensitivity requests to older models, triggering a wave of criticism.

Safety Built In, Not Bolted On

In an apparent effort to distance itself from the criticism Anthropic faced, OpenAI emphasized that Sol's safety measures are embedded directly into the model's core behavior — not added as an external filter layer that can be bypassed or swapped out. The company also designed the model to be more resistant to jailbreak attempts, and has oriented its cybersecurity capabilities toward defense rather than offensive exploitation.

What's Next?

OpenAI plans to make GPT-5.6 more broadly available through ChatGPT, Codex, and its API in the coming weeks, following the completion of necessary reviews. The current landscape makes clear that the real battle is not merely between competing tech companies — it extends to a fundamental redefinition of roles between governments and AI developers in determining who controls the future of this technology and the pace at which it spreads.

✦ بقلم فريق دروب أيديا