Trump's AI order now makes model vetting voluntary instead of mandatory, representing a major reversal in the administration's frontier AI governance approach.
The latest version of Trump's planned AI executive order would give frontier model developers a 90-day window to participate in optional government vetting, according to Gizmodo. This marks a retreat from earlier iterations that would have imposed mandatory review through a dedicated federal agency. The order splits oversight into two tracks: cybersecurity infrastructure hardening and model safeguards, though the latter carries no binding requirements for participation.
Meanwhile, elsewhere in the industry, companies are moving decisively forward with AI deployment and monetization strategies. Pharmaceutical firms like Bristol Myers are rolling out Claude to tens of thousands of employees to accelerate drug development, while the broader sector navigates complex business model pressures. The contrast between this aggressive expansion and Washington's opt-in regulatory approach suggests a widening gap between industrial reality and policy intent.
What emerges from these parallel developments is a fundamental puzzle: how does optional compliance with AI safeguards actually constrain anything? This reporting explores what happens when an industry moving at scale confronts a regulatory approach based entirely on voluntary participation. The disconnect between these two forces may ultimately determine the future of AI governance.
The Two-Part Structure: Cybersecurity and Voluntary Frameworks
On May 20, 2026, gizmodo.com reports that the Trump administration's forthcoming AI executive order divides into two distinct sections: cybersecurity infrastructure hardening and a separate covered frontier models section. The cybersecurity component aims to strengthen federal defenses against perceived AI-related threats while avoiding direct regulation of AI development itself. Rather than constraining model development, this section addresses what the administration views as critical infrastructure vulnerabilities associated with advanced artificial intelligence systems.
The covered frontier models section takes a notably permissive approach. mashable.com noted that recent announcements of advanced models like Gemini 3.5 and multimodal Gemini Omni exemplify the frontier systems potentially subject to the framework. Under the executive order, makers of such frontier models would have a voluntary 90-day window to check in with the government and have their model vetted, but participation remains entirely optional rather than mandatory, creating questions about practical incentives for companies to engage.
The two-track structure reflects a deliberate compartmentalization of concerns. Cybersecurity hardening allows infrastructure protection without constraining commercial innovation in model development, while the voluntary frontier models framework preserves industry freedom even as it creates a channel for optional government engagement. This approach maintains the administration's fundamental deregulatory orientation while creating minimal oversight mechanisms.
A Remarkable Policy U-Turn in Six Months
Approximately two months prior to this May 2026 proposal, the Trump administration had signaled a markedly different approach to AI governance. gizmodo.com reports that an earlier policy document released around March 2026 emphasized minimal regulation, proposing only age restrictions for users rather than any framework governing model development or industry oversight. The administration's broader deregulatory perspective had been publicly articulated in early 2025 by Vice President J.D. Vance during a speech at France's AI Action Summit, where he essentially rejected AI regulation altogether.
Recent announcements indicate the frontier model landscape has evolved dramatically, potentially shifting administration priorities. techcrunch.com reported that at Google's May 2026 I/O event, the company unveiled advanced new models including Gemini Omni, a multimodal system with cloud-based agent capabilities reaching hundreds of millions of users globally. The acceleration of frontier model development and deployment across industries may have prompted the administration to reconsider its original hands-off stance.
The chaotic policy drafting revealed by Axios sources reflects significant internal tensions about AI governance. What began as emphatic rejection of AI regulation by the vice president evolved into a structure offering government engagement pathways with frontier models, even if voluntary. This shift suggests the Trump administration has grappled with balancing its original deregulatory commitments against perceived needs for at least minimal oversight mechanisms as frontier capabilities advance.
The Frontier Model Launch Frenzy Meanwhile
Google unveiled its latest AI capabilities at I/O 2026 with TechCrunch reporting that Gemini 3.5 Flash now powers the default model in both the Gemini app and Google Search's AI Mode, delivering output at four times the speed of competing frontier models. The company simultaneously announced Gemini Omni, a multimodal world model that processes text, audio, images, and video inputs to generate scientifically grounded content, with DeepMind CEO Demis Hassabis framing it as a pivotal step toward artificial general intelligence. Both models launched immediately to paid subscribers, signaling Google's accelerated deployment strategy in the race for market dominance. These releases represent a fundamental shift in how frontier AI capabilities reach users, moving from research announcements to immediate, widespread availability.
Meanwhile, Mashable highlighted Google's introduction of Gemini Spark, described as a 24/7 personal AI agent that operates continuously in the cloud and integrates with over 30 third-party tools including Adobe, Dropbox, and Uber through MCP standards. Rolling out first to Google AI Ultra subscribers in Gmail and Chat within a week, Spark demonstrates the shift from conversational assistants to autonomous agents capable of executing workflows independently. The agent can independently pull emails, files, and calendar data to compile executive briefings without human intervention at each step. This agentic approach contrasts sharply with traditional chatbot deployments, establishing a new performance baseline for enterprise AI applications.
Concurrently, TheNews.com.pk reported that Anthropic's Claude is now deployed across over 30,000 Bristol Myers Squibb employees specifically for accelerating drug discovery and development workflows. Bristol Myers is simultaneously exploring integration of Claude Code, Anthropic's coding tool, into research and commercial operations, moving beyond simple chatbot deployments toward domain-specific applications in pharmaceutical development. The deployment reflects pharmaceutical companies' growing confidence in using frontier models for specialized knowledge work. These parallel launches underscore how frontier model providers are competing not just on capability announcements but on real-world deployment velocity and integration breadth.
The Credibility Gap: Voluntary Frameworks and Industry Momentum
According to Gizmodo, the Trump administration's latest version of its frontier AI executive order establishes a voluntary vetting framework rather than mandatory oversight, allowing model makers a 90-day window to optionally submit their systems for government review. The shift represents a significant departure from earlier drafts that appeared to position a federal Center for AI Standards and Innovation as the gatekeeper for all frontier model releases, though the voluntary mechanism raises immediate questions about enforcement and incentive structures. As Gizmodo's analysis notes, a voluntary framework doesn't align with sound business logic when companies can simply skip the process without legal consequence. The decision reflects internal administration tensions around how aggressively to regulate AI versus maintaining alignment with an industry that strongly resists prescriptive government controls.
Yet the industry's actual deployment pace suggests that market forces rather than government frameworks are shaping AI rollout strategies. TechCrunch and Mashable both documented how Gemini Spark and Gemini Omni rolled out to production within days of announcement, while agentic AI systems are reaching enterprise deployments with minimal delay. Companies are racing to embed multimodal models and autonomous agents into core business operations from pharmaceutical research to personal productivity on timelines measured in weeks rather than months. The competitive pressure to match OpenAI's and Anthropic's feature velocity leaves little room for extended regulatory review cycles, even voluntary ones.
This divergence creates a credibility problem for any oversight framework that cannot keep pace with deployment cycles. When frontier models reach thousands of employees and integrate with dozens of third-party services faster than government bodies can complete vetting processes, the distinction between voluntary and mandatory oversight becomes largely academic. The industry's momentum toward multimodal and agentic AI appears driven primarily by technical capabilities and competitive positioning rather than by regulatory guardrails, suggesting that government frameworks may need fundamentally different enforcement mechanisms to meaningfully shape industry behavior rather than simply formalize what companies already intend to do anyway.
The Regulatory Whiplash Behind Model Sharing
The Trump administration's shift from mandatory government vetting to a voluntary framework represents a stunning reversal from its own earlier positioning. Vice President J.D. Vance publicly renounced AI regulation entirely in a 2025 speech in France, yet the emerging executive order now proposes structured oversight of frontier models, however loosely. This pivot suggests internal pressure,from security advisors, industry stakeholders, or both,to address legitimate concerns about unvetted advanced AI systems, even within an administration ideologically opposed to regulation.
The voluntary framework creates a perverse incentive structure that remains unexamined in coverage. If AI developers can opt out of a 90-day government vetting process without consequences, the Gizmodo report itself notes this "doesn't make much business sense." Without penalties for non-participation or tangible benefits for compliance, the framework collapses into performative safety theater,companies gain reputational cover by submitting models while others skip the process entirely. The real teeth of regulation lies in enforcement mechanisms the order apparently lacks.
This moment reveals a deeper tension: AI companies are racing to monetize frontier models at scale (as evidenced by OpenAI's ChatGPT Go rollout and enterprise deals like Bristol Myers' Claude deployment) while governments worldwide grapple with how to govern them. A voluntary framework satisfies neither genuine safety oversight nor industry freedom,it signals compromise under pressure, not coherent policy.
The Trump Administration's executive order on AI has undergone a marked transformation from its initial conception. What appeared to be a move toward comprehensive government oversight of frontier AI models has shifted toward a voluntary framework, where companies have 90 days to voluntarily submit models for vetting. This reversal reflects ongoing tensions within the administration itself, with earlier deregulatory impulses giving way to a more cautious approach, even as that approach stops short of mandatory requirements. The result is a policy halfway between the hands-off stance of 2025 and the stricter proposals of weeks past.
Meanwhile, the private sector has continued its relentless pace of product launches, partnerships, and capability expansions. OpenAI, Google, and Anthropic are simultaneously scaling their AI services, building new product categories, and integrating their models deeper into commercial workflows. This parallel trajectory suggests that regardless of whether government frameworks are voluntary or mandatory, the momentum of private AI development may render the choice largely academic. The real question facing policymakers is not whether they can slow the market, but whether they can design oversight structures that remain relevant even as the technology outpaces the policy.
Frequently Asked Questions
What did Trump's AI executive order originally propose?
The order initially suggested creating a government agency like the Center for AI Standards and Innovation to vet all new frontier AI models, a significant shift from the administration's 2025 deregulatory rhetoric.
Will companies have to follow the new AI rules?
No. The framework is voluntary, though companies have 90 days to check in with the government, and compliance is entirely optional.
Why did the Trump administration change its AI policy?
Leaked reports suggest internal disagreements within the administration, leading to a reversal from an extremely permissive stance toward a more cautious voluntary approach.
Are AI companies slowing down while waiting for regulations?
Not visibly. Major providers like OpenAI, Google, and Anthropic continue launching new products, models, and enterprise partnerships at an accelerating pace.
Could a voluntary framework actually protect against AI risks?
Uncertain, since companies can ignore vetting recommendations, which raises questions about whether voluntary measures can achieve meaningful safety outcomes without enforcement mechanisms.