Mythos 5 & Fable 5 Curbs Lifted: Inside Trump’s ‘Minimal Guardrails’ AI Shift

Last month, the AI research world was jolted when the U.S. government forced Anthropic to shut off access to its most capable models—Mythos 5 and Fable 5—only to walk back the controls weeks later. This whiplash illustrates the operational reality behind Trump’s slogan of minimal regulation: guardrails exist, but they are applied with surgical speed and then removed just as quickly. For AI developers and researchers, the message is clear: the ability to deploy cutting-edge models can hinge on classified security assessments and diplomatic backchannels, not just code.

On July 2, 2026, President Donald Trump articulated his administration’s core philosophy on artificial intelligence regulation with characteristic bluntness: “Well, you need some guardrails, but you want to do as little as possible.” The comment, made during a CNBC interview at the White House, arrives just weeks after an extraordinary sequence of events that tested that very principle. The Commerce Department abruptly imposed export-style controls on Anthropic’s frontier models, Mythos 5 and Fable 5, compelling the company to shut them down and enter urgent negotiations—only for the administration to partially reverse course within days. The episode, and Trump’s sidestep of questions about a proposed 5% U.S. government equity stake in OpenAI, offers a rare glimpse into how the world’s leading AI power is balancing national security, industrial competition with China, and a deeply ingrained deregulatory impulse.

“

government forced Anthropic to shut off access to its most capable models—Mythos 5 and Fable 5—only to walk back the controls weeks later.

The Anthropic saga represents the most concrete demonstration yet of the administration’s willingness to intervene when it perceives immediate cybersecurity threats. In mid-June 2026, the Commerce Department imposed curbs requiring Anthropic to seek U.S. permission before granting any foreign national access to Mythos 5 or Fable 5. The rationale cited cybersecurity concerns, though details remained classified. The move stunned Silicon Valley, where Trump’s AI policy had been widely interpreted as laissez-faire, especially compared to the EU’s prescriptive AI Act. Anthropic disabled the models, and back-channel talks began. By the last week of June, the government eased restrictions on Mythos 5, restoring access for pre-approved U.S. organizations. Then, in the first days of July, it fully lifted the curbs on Fable 5. The rapid about-face underscores the reactive, case-by-case nature of the approach that Trump endorsed.

This pattern—aggressive intervention followed by quick unwinding—creates a profoundly uncertain regulatory environment for AI companies. On one hand, the president’s statement explicitly frames minimal rules as a competitive advantage against China: burdensome restrictions, he implied, could hamper American firms. On the other, the Anthropic example shows that even a “light-touch” administration will not hesitate to shut down access to powerful models if it perceives a risk. The message for startups and established players alike is that the guardrails are real but unpredictable, and that the threshold for action can be crossed with little warning.

The OpenAI equity proposal adds another layer of complexity. Citing a Financial Times report, Trump was asked about preliminary discussions in which OpenAI chief Sam Altman had proposed giving the U.S. government a 5 percent stake. The president sidestepped, instead touting a 2025 deal that gave Washington a stake in Intel. If the OpenAI proposal progresses, it would mark a novel model of government involvement: an ownership position designed to align interests and potentially give the state a louder voice in strategic decisions without formal regulation. For AI firms, the prospect of the government as a shareholder introduces both opportunities (political alignment, preferential treatment) and risks (loss of independence, future demands).

The competitive dimension with China is the unifying thread. Trump’s “as little as possible” framing is explicitly tied to ensuring that American AI companies are not burdened while China accelerates. Yet the Anthropic controls were export-oriented—they were about restricting foreign access, exactly the kind of measure that might be applied to prevent Chinese entities from obtaining advanced capabilities. This duality—minimal domestic red tape, maximal vigilance on foreign leakage—could define the next phase of U.S. AI policy. Companies that develop dual-use or exceptionally capable models should anticipate that the line between national security and commercial freedom can be redrawn quickly.