The U.S. government is poised to take unprecedented control over which AI models reach the public, fundamentally reshaping the competitive landscape. This shift, intensifying in 2026, means the rivalry between Anthropic and OpenAI is no longer the central story.
Just two weeks after the U.S. government pulled Anthropic’s Fable and Mythos models, OpenAI's latest system faces a similar fate. The Information reported Thursday that GPT-5.6 will launch only in a limited preview, with government approval required on a “customer by customer” basis before any general release is permitted.
If this preview lasts just a “couple of weeks,” as OpenAI CEO Sam Altman reportedly projected, the impact might be manageable. But Anthropic’s Mythos has been stuck in preview for months with no clear path to full release. Even a short review period could severely limit the economic upside of expensive new systems, at a time when AI labs desperately need to improve their bottom lines. A slowdown in model development would likely also chill the ongoing data center buildout, which has been a major economic driver in 2026.
If this situation deteriorates, the entire industry could face existential risk.
Critically, OpenAI and Anthropic now face identical challenges: the same regulatory hurdles, the same potential for disaster if they fail. Tech industry conversations often blame one side or the other—accusing Anthropic of regulatory capture or OpenAI of cozying up to political powers to thwart rivals. This is understandable, given the billions at stake. However, what's unfolding transcends corporate rivalry.
The cost of a haphazard government approval process for every frontier model is clear, and no solution that helps one lab can avoid helping others. The most immediate issue is establishing a sensible release framework. Government testing of models before release is not inherently problematic—it mirrors processes for many consumer products. Yet, as GMU fellow and soon-to-be OpenAI employee Dean Ball detailed recently, regulators lack the expertise and capacity for the necessary evaluations. It remains unclear what safety assurances would satisfy them, or even what specific risks the government aims to mitigate.
While the flawed approval process is a major concern, genuine risks exist. Even skepticism about the Mythos hype doesn't negate clear evidence that AI tools are revolutionizing cybersecurity, with similar implications for biorisk and alignment. Restricting model releases can't be the only solution—it merely limits public access—but the underlying issues demand attention.
As Ball argues, the best way forward requires collaboration: trusting independent groups to guide the process, even when their goals differ from one's own; uniting behind the least-bad regulatory options instead of fighting every rule; and, most importantly, advocating for AI as an industry rather than using safety and regulation as competitive weapons.
via TechCrunch AI