Breaking

OpenAI Restricts Access to New GPT-5.6 Models After Talks With Administration

June 26, 2026 at 14:29 EDT

Foundation Models
Policy & Regulation

In June 2026, OpenAI said it would limit access to its next-generation model family "GPT-5.6" following discussions with the Trump administration. Initial access is confined to a small set of government-approved partners, with a staggered rollout.

June 25, 2026 · OpenAI

Government Gatekeeping: GPT-5.6 Released Only to Approved Partners

After talks with the Trump administration, OpenAI will limit its newest model to a small set of government-approved companies — cleared customer by customer — over concerns its cybersecurity capabilities reach a level warranting prior review.

1-by-1

Access approved customer by customer, with phased expansion to follow

1.5M

Tokens — reported expanded context window, with stronger agentic performance

≤30 days

Pre-release window sought by a June executive order — framework not yet in place

API output pricing — reported tiers

Price per 1M output tokens · positioned cheaper than comparable rival models

$30

Sol

$15

Terra

Luna

How we got here

Anthropic models halted

Commerce directive stops Mythos & Fable

→

GPT-5.6 judged "Mythos-like"

Comparable cybersecurity capabilities

→

Access restricted

Approved partners only; public release weeks out

Reasonable safeguard

Some see prior government approval as a sensible national-security measure amid competition with Chinese open-source models and leakage risk to nation-states and cybercriminals.

Innovation at risk

Others warn of slowed innovation and ad-hoc regulation becoming the norm. Even Altman's memo calls the customer-by-customer setup "not desirable" as a long-term model.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.