US quietly reopens the AI gate: Anthropic and OpenAI get vetted access—what does it mean for regulation and cyber power?

The Trump administration has moved to ease restrictions around advanced AI access, allowing some use of Anthropic’s “Mythos” after a prior two-week lockout. Reporting indicates the US government permitted Anthropic to share an AI model with selected firms, signaling a shift from blanket controls toward case-by-case approvals. Separately, OpenAI announced a “limited preview” of GPT-5.6 for users vetted by the US government, with emphasis on powerful cyber security capabilities. Together, the two announcements suggest Washington is actively shaping who can deploy frontier models and for what purposes, rather than relying on a single, stable regulatory framework. Strategically, this is a geopolitical signal about AI governance as a tool of national power. By vetting recipients and controlling distribution channels, the US can steer model deployment toward sectors that align with security priorities, while limiting proliferation to actors deemed higher risk. The tension highlighted by the Financial Times—unease over Washington’s ad hoc regulatory approach—implies uncertainty for industry planning and could intensify lobbying for clearer rules. In practice, the policy benefits the most connected firms able to pass government scrutiny, while disadvantaging smaller labs and non-vetted customers that face slower access or uncertain compliance pathways. Market and economic implications are likely to concentrate in AI infrastructure, cybersecurity services, and enterprise software procurement. If GPT-5.6’s cyber capabilities translate into measurable reductions in breach risk, demand could rise for managed security, identity protection, and incident-response tooling, supporting revenue for vendors tied to cloud security stacks. For Anthropic, renewed access to “Mythos” may improve competitiveness in enterprise deployments and accelerate partnerships with selected firms, potentially affecting model-choice dynamics among large customers. In financial markets, the immediate impact is less about broad index moves and more about relative positioning: investors may re-rate cybersecurity-exposed AI beneficiaries and those with government access pathways, while penalizing firms facing regulatory friction. What to watch next is whether the US formalizes the vetting criteria, expands the list of approved recipients, or introduces additional constraints tied to specific model capabilities. Key indicators include the duration of any future moratoria, the transparency of approval standards, and whether “cyber security capabilities” become a defined compliance category. Another trigger point is whether other frontier labs seek similar access and whether regulators respond with consistent timelines or continued ad hoc decisions. Over the next weeks, market participants should monitor announcements of additional previews, changes in user eligibility, and any signs that model distribution is being linked to measurable security outcomes or audits.

Geopolitical Implications

• 01

  AI distribution control is becoming a form of strategic governance, allowing the US to align frontier model deployment with national security priorities.

• 02

  Selective access may widen the gap between well-connected US-aligned firms and smaller labs, shaping the competitive landscape of frontier AI.

• 03

  Ad hoc regulation can trigger lobbying and policy churn, potentially affecting cross-border AI collaboration and raising compliance costs for global customers.

• 04

  Cyber-focused model capabilities may strengthen US influence in security ecosystems, with downstream effects on incident-response and defense-adjacent procurement.

Key Signals

• —Any publication or clarification of vetting criteria and approval timelines for frontier AI models
• —Announcements of additional model previews or expansions of the approved user list
• —Whether future restrictions target specific capabilities (e.g., cyber, dual-use, autonomy) rather than the model itself
• —Customer adoption signals from selected firms using Mythos or GPT-5.6 in security workflows