White House AI and Crypto Czar David Sacks on Monday, 20 July 2026 publicly criticised American AI models for refusing to fix security vulnerabilities on safety grounds, arguing that Chinese counterparts are completing the same tasks without restriction and that US competitiveness is suffering as a result.

Context

In his post on X, Sacks wrote that Kimi K3 — a large language model developed by Chinese AI company Moonshot AI — 'just fixed 15 critical security bugs that Codex and Fable refused because of cyber guardrails.' He added: 'There's no reason to limit American models on tasks that Chinese models handle without issue. We're only making ourselves less competitive.'

The statement is a direct broadside at the content-restriction policies built into frontier American AI coding tools. Codex, developed by OpenAI, is among the most widely used AI systems for code-related tasks and has publicly documented guardrails around potentially dangerous or dual-use outputs.

Policy Backdrop

The debate Sacks is wading into has a long policy lineage. The Biden administration's Executive Order on Safe, Secure, and Trustworthy AI, issued in October 2023, directed federal agencies to develop standards for AI safety testing and red-teaming — frameworks that influenced how US labs calibrated model refusals on security-adjacent tasks.

Separately, the US Department of Commerce has progressively tightened export controls on advanced AI chips and model weights to China since October 2022, a policy designed to slow Chinese AI development. Critics of that approach — and of domestic safety guardrails — argue both sets of restrictions are self-defeating if Chinese labs face lighter oversight at home and continue shipping capable models globally.

The tension between national-security rationales and commercial innovation priorities has resurfaced repeatedly: in debates over open-source model releases, chip controls, and now model-level content policies for cybersecurity use cases.

Stakeholders and Impact

US AI developers and cybersecurity researchers sit at the centre of this dispute. Security professionals who rely on AI-assisted vulnerability discovery and patch generation face a direct operational consequence if American models decline tasks that foreign alternatives complete. For enterprise and government clients, model refusals on legitimate security workflows can push procurement toward non-US systems.

For Moonshot AI, Sacks's post — whether or not every specific claim is independently verifiable — amounts to high-profile Western validation of its Kimi model family's capabilities, at a moment when Chinese AI labs are actively competing for global developer mindshare.

The comment also carries institutional weight given Sacks's role. As the Trump administration's designated AI and crypto policy lead, his public statements signal the direction of forthcoming executive guidance on how the US government views model safety constraints relative to competitiveness imperatives.

What's Next

Observers are watching for any forthcoming Trump administration AI executive order or guidance from the Office of Science and Technology Policy (OSTP) on security-related model evaluations. If Sacks's framing prevails in internal deliberations, US labs could face pressure — formal or informal — to revise their red-teaming and refusal policies for cybersecurity tasks.

The broader implication is significant: if Washington concludes that voluntary safety guardrails are eroding American AI leadership, the regulatory mood could shift from encouraging restraint to penalising it — a reversal that would reshape how frontier models are built and deployed across the industry.