Skip to main content
Return to Blog de Seguridad de TrendAI™
AI & emerging technologies APTs Compliance

What the Blue Wave Means for AI Regulation

Congress is poised to set new AI rules. Tom Kellermann outlines five areas to watch, from AI agent governance to workforce preparation, and what defenders can do now.

Perspectives AI governance APTs Data privacy & regulation

On November 3, 2026, voters gave Democrats control of the House and the Senate, and they sent a message about AI regulation along with it. More than half of voters said in an NBC poll that AI and data centers played a crucial or major role in their vote for Congress.

For the past two years, federal AI policy has been thin or non-existent, and states have filled the gap on their own . An executive order on December 11, 2025 even set up a DOJ AI Litigation Task Force to challenge state AI laws in federal court.

A post-election tally by Data Center Watch found the vast majority of new governors and members of Congress had called for tougher rules in their campaigns. The new blue wave may not stop the White House from challenging AI regulation, but I predict we'll see Congress set a new course in five key areas:

  1. Governance and control standards for AI agents

This is the one I care about most. Ungoverned AI should be viewed as command and control for an adversary. It is a new living-off-the-land technique. Companies are handing agents access to email, cloud environments and internal tools with less scrutiny than a new hire gets, and every one of those agents is like an insider with no background check. TrendAI™'s Global AI Study found 67% of IT and business decision makers said they had felt pressured to approve AI despite security concerns.

Expect Congress to push a standard of care for guardrails that defines what reasonable governance of an agent looks like, aligned to the NIST AI Risk Management Framework : a governance gateway so you know where your AI begins and ends, continuous monitoring of agent behavior rather than agent permissions, and triggers that fire when someone attempts to break out of the guardrails. Zero trust alone won't get you there. You need to continuously verify before you trust, because agents carry implicit privilege. A NIST-aligned standard gives CISOs the leverage to demand those controls from vendors and from their own business units.

2. A frontier model incident-reporting mandate

According to TrendAI™’s H1 2026 APT Activity Roundup, nation-state APT actors are already “riding the dragon,” or using generative models to automate reconnaissance, lateral movement, and exploit generation to compress days’ worth of attack steps into minutes. When a model is jailbroken at scale or abused in a campaign, the rest of us need to know while the intelligence is still actionable.

Expect a requirement that frontier model developers report serious safety and security incidents within a 36-to-72-hour window, modeled on the Cyber Incident Reporting for Critical Infrastructure Act of 2022 (CIRCIA)’s 72-hour clock and banking regulators’ 36-hour rule .

3. Government oversight of frontier model access

Today, access to the most capable models is governed by a vendor’s terms of service. But the capabilities that make these models valuable to defenders are the same ones adversaries are after, and right now the vendor alone decides who gets them.

Expect know-your-customer obligations for model providers, tiered access with stricter vetting for the most capable systems, and requirements that firms in critical infrastructure, finance, and defense demonstrate governance and security controls before they can deploy frontier models in production.

4. Data center environmental and energy standards

Data centers were a visible AI-related ballot issue in many places, with voters wanting a say in how buildouts occur in their communities.

Expect federal standards on energy use, water consumption, and grid impact, along with reporting requirements so communities can see what they are supporting and what they get in return.

5. Preparing the workforce

Blue Rose Research polling from March 2026 found that 77% of Americans are concerned about entire industries being eliminated by AI faster than new ones can be created, and 79% worry the government has no plan to protect workers from the fallout.

Expect serious money for AI skills programs, and expect cybersecurity to be named in them. We need threat hunters who can hunt agents, incident responders who expect the adversary to fight back, and compliance teams trained to treat AI agents the same way they would treat an insider who’s a known criminal. Offense informs defense. We should be training for the fight we're actually in.

What this means for defenders

None of this will be law tomorrow, but my advice: don't wait for the mandate. Figure out where your AI ends and begins. Treat every ungoverned agent as hostile infrastructure. Map your attack paths continuously and assume the adversary is already inside. Organizations that get their AI under control now will already be in compliance when these rules arrive. Everyone else will be scrambling.