The Great American AI Act Draft: What Security Leaders Should Do Before It Becomes Law

On June 4, 2026, Reps. Lori Trahan (D-Mass.) and Jay Obernolte (R-Calif.) released a discussion draft of the Great American AI Act, a nearly 270-page bipartisan attempt to build a single federal framework for how the United States governs artificial intelligence. It is not law yet. It is not even formally introduced. But for security and governance leaders, a discussion draft is exactly the moment to pay attention, because the direction it sets will shape compliance obligations, vendor expectations, and security program priorities well before any final vote.

The draft arrived two days after the White House issued a scaled-back AI executive order, and it pulls together threads that have been scattered across a patchwork of state laws and executive actions. Below is what is actually in it, and more importantly, what you should be doing about it now.

What the Draft Actually Proposes

The bill is sprawling, but a handful of provisions matter most for cybersecurity and AI security teams:

• A codified federal standards center. The draft would formally establish the Center for AI Standards and Innovation (CAISI) within the Commerce Department, with $100 million authorized per year for fiscal 2027 through 2029. CAISI would develop voluntary guidelines, best practices, and standards for AI security, evaluate AI systems, and monitor AI progress. The center currently exists (rebranded from the Biden-era AI Safety Institute) but lacks formal authorization from Congress.
• Frontier incident reporting. Large frontier model developers would be required to report critical safety and security incidents to the government, according to Politico.
• Three-year state preemption. The draft would preempt states from passing their own laws regulating the development of frontier AI models for three years, while still allowing laws of “general applicability” and regulation of models after deployment, as Nextgov reported.
• Open-source security funding. The Cybersecurity and Infrastructure Security Agency would be authorized to award grants to maintainers of widely used open-source software for patching, maintenance, and security audits. The GAO would evaluate the security of the open-source ecosystem and the protocols protecting AI model weights.
• Extension of threat-sharing authority. The draft would extend the Cybersecurity Information Sharing Act of 2015 through fiscal 2035, per GovTech.
• New penalties and testbeds. It adds penalties for using AI to impersonate government officials, directs the Energy Department, NIST, and NSF to build an AI evaluation testbed program, and includes whistleblower protections and anti-fraud provisions, as noted by MeriTalk and Axios.

Why This Matters Even Though It Is Just a Draft

It is tempting to file a discussion draft under “watch and wait.” That would be a mistake for three reasons.

First, the security-relevant provisions reflect where bipartisan consensus is forming: model weight protection, frontier incident reporting, open-source supply chain security, and federal standards. Those themes are unlikely to disappear even if this specific bill changes shape.

Second, the preemption clock creates urgency. If a federal framework preempts state frontier AI laws for three years, organizations that have been building compliance programs around individual state requirements will need to reconcile them against a single federal baseline. The companies that mapped their obligations early will adapt faster.

Third, voluntary standards have a way of becoming de facto mandatory. CAISI guidelines will shape procurement language, insurance requirements, and contractual expectations long before they carry the force of law. The NIST AI Risk Management Framework already follows this pattern.

What to Do Now

1. Map Your AI Systems Against Frontier Definitions

The incident-reporting and preemption provisions hinge on what counts as a “frontier” model. Even if you are not training frontier models, you almost certainly consume them through APIs and vendors.

• Inventory every AI system you build, fine-tune, or integrate, including third-party APIs and embedded model features.
• Note which providers would likely fall under frontier developer obligations, since their new reporting duties may surface incident information you can use.
• Flag systems where you are the deployer rather than the developer, because post-deployment regulation would still apply to you even under preemption.

2. Build an AI Incident Reporting Capability Before It Is Required

If frontier developers must report critical safety and security incidents, downstream organizations will face pressure to do the same through contracts and customer expectations.

• Define what a “critical AI incident” means for your organization (model compromise, prompt injection leading to data loss, unsafe autonomous action, training data poisoning).
• Extend your existing security incident response plan to cover AI-specific failure modes, rather than building a separate process.
• Establish who is accountable for detecting, triaging, and reporting AI incidents, and rehearse it.

3. Treat Model Weights as Crown-Jewel Assets

The GAO would be directed to evaluate the security protocols protecting AI model weights. That signals where federal scrutiny is heading.

• If you train or fine-tune proprietary models, classify the weights at your highest data sensitivity tier.
• Apply access controls, encryption, and monitoring to weight storage and transfer the same way you would to source code or cryptographic keys.
• Audit who can export or copy model artifacts, and log it.

4. Get Ahead of Open-Source AI Supply Chain Risk

The CISA grant program and GAO open-source evaluation underscore that the software supply chain remains a top-tier concern, and AI has expanded it dramatically.

• Maintain a software bill of materials that includes AI libraries, model dependencies, and datasets.
• Track the maintenance health of the open-source AI components you rely on, since unmaintained packages are a known attack surface.
• Apply the same vulnerability management discipline to AI dependencies that you apply to the rest of your stack.

5. Consolidate Your Compliance Mapping

If preemption arrives, a federal baseline will eventually replace the state-by-state approach for frontier development.

• Build a single control framework that maps to NIST AI RMF, anticipated CAISI guidance, and the state requirements you currently track.
• Avoid hard-coding compliance to any one state regime that could be preempted.
• Keep the mapping living, because a discussion draft will change before it becomes law.

6. Submit Feedback

This is a discussion draft expressly soliciting input, with comments directed to GAAIA@mail.house.gov. Security and governance practitioners have a rare opportunity to shape definitions and reporting thresholds before they harden into statute. If incident-reporting timelines or frontier thresholds would be impractical to operationalize, now is the moment to say so.

The Bottom Line

The Great American AI Act draft is a signal, not a mandate. But the signal is clear: federal AI governance is converging on model weight protection, frontier incident reporting, open-source supply chain security, and a single national standard. Security leaders who treat this draft as a planning prompt, mapping their AI inventory, extending incident response to AI failure modes, and hardening model artifacts, will be ready regardless of how the legislative process unfolds. Those who wait for a signed law will be reacting to deadlines instead of setting their own pace.

The strongest cybersecurity and AI security programs do not wait for regulation to tell them what good looks like. They use drafts like this one to see where the puck is going and skate there first.

---

Sources: FedScoop; supporting reporting from Nextgov, Politico, Axios, GovTech, and MeriTalk. The official discussion draft and section-by-section summary are available from the office of Rep. Jay Obernolte.