Washington Won't Regulate AI. The States Are Trying, and Washington Is Suing Them Anyway.

Attorney General Todd Blanche has drawn the clearest line yet around the federal government's role in artificial intelligence: the Justice Department will prosecute crimes that happen to involve AI, but it will not regulate the technology itself. The distinction, laid out in a September interview with Bloomberg News, has created a vacuum at the center of American AI policy. And the states, Washington, and the courts are all fighting over who gets to fill it.

"I'm not going to do regulation by prosecution and investigate or prosecute AI companies when there's not a statute that they're violating," Blanche said. "If we're going to charge somebody with violating the laws and take away their liberty and put them in prison, it should be because they violated the law."

A deliberate break from the Biden era

Blanche, confirmed as attorney general in August 2026, framed the stance as a correction. He accused his Biden administration predecessors of spending "countless prosecutors' hours and money and effort and resources" using enforcement as a substitute for legislation in the technology sector. "There are some Justice Departments that like to do that, and that's not what we're doing," he said.

The policy has a pedigree. In April 2025, the department issued a memorandum declaring it "is not a digital assets regulator," directing prosecutors to stop pursuing crypto cases that effectively imposed regulatory frameworks on the industry. A Bloomberg Law analysis published Oct. 16 by Abhishek Kambli, a former deputy associate attorney general, traces the AI stance directly to that playbook: the same logic, the same language, now governing artificial intelligence.

The administration then formalized the posture. A December 2025 executive order established an AI Litigation Task Force whose stated mission is "to sustain and enhance the United States' global AI dominance through a minimally burdensome national policy framework for AI." The federal role, as defined, is protecting the industry's competitive position, not building an oversight architecture. Appearing on Fox & Friends Weekend in late September, Blanche dismissed calls for new AI safety measures outright: "Everybody's saying more laws, more laws, more regulation, but we don't need it."

The federal government has defined its role as protecting the AI industry's competitive position, not building an oversight architecture. The states are writing rules anyway, and Washington is now suing to stop them.

States fill the gap

With federal hands off, states have moved to write their own rules. California's Transparency in Frontier AI Act, effective this year, requires large frontier model developers to publish information about training data, safety frameworks, and risk assessments, with penalties reaching $1 million per violation. New York, Illinois, and Connecticut have enacted similar statutes. Illinois mandates annual independent third party audits of safety practices. Connecticut requires bias testing and pre-use notices when AI tools are used in employment decisions.

State attorneys general are also testing existing law against AI companies. Pennsylvania's attorney general brought an enforcement action against a chatbot that falsely represented itself as a licensed psychiatrist. Florida's attorney general opened a criminal investigation into OpenAI's ChatGPT after the chatbot allegedly offered advice to a gunman. In August, 15 state attorneys general demanded that OpenAI preserve evidence related to its products, a sign of how seriously state enforcers now take AI incidents.

The state level is where the sharpest new theories are emerging. On Sept. 8, Florida's attorney general proposed legislation that would hold companies liable when their AI systems are involved in criminal activity. Rather than asking whether a company's executives knowingly broke the law, the proposal asks whether the product contributed to someone else breaking the law. It is a fundamentally different theory of corporate responsibility, and it is being built entirely outside Washington.

Washington sues the substitutes

The federal government is not merely stepping aside. It is actively contesting the state level rules that fill the void. In April 2026, xAI sued to block Colorado's Artificial Intelligence Act (SB24-205), the country's first comprehensive algorithmic discrimination law. Days later, the Justice Department intervened in support of xAI, the first time the federal government had moved to invalidate a state AI law. The department argued the statute violated the Equal Protection Clause by forcing developers and deployers of AI systems to monitor and adjust their outputs for discriminatory patterns across protected characteristics, and raised a First Amendment argument about compelled speech.

Colorado responded by rewriting the law. The revised statute, recast as a transparency focused Automated Decision Making Tools (ADMT) Act, drops the algorithmic discrimination provisions entirely and requires only pre-use notices, plain language disclosures after adverse outcomes, and opportunities for human review. The equal protection argument was neutralized by the rewrite, though the compelled speech question may survive.

As of October 2026, the revised Colorado law is formally effective Jan. 1, 2027, but enforcement is stayed by the federal district court in xAI v. Weiser until the court rules on xAI's preliminary injunction motion. The Colorado attorney general filed proposed implementing rules on Aug. 11, with the public comment period running through Oct. 26, 2026. If the DOJ's challenge succeeds, it could narrow the field of state laws the federal government is simultaneously declining to replace.

The accountability gap

The combined effect is a fragmented accountability landscape. The DOJ will not regulate AI through enforcement. The FTC is pursuing a separate track through civil investigative demands to OpenAI and Anthropic, shifting from marketing claims to structural safety discovery. Congress, as of October 2026, has not passed a federal AI statute: the Warner AI AGENT Act and the Hawley-Murphy AI Agent Accountability Act both remain in early legislative stages with no committee markup.

Existing federal criminal law was built for human hackers, not autonomous agents. The 40-year-old Computer Fraud and Abuse Act makes it illegal to knowingly access a computer without authorization, and the White House has directed prosecutors to use it against those who employ AI to illegally access systems. But former DOJ officials cited in an Oct. 4 Associated Press report note that the statute's "knowingly" and "intentionally" requirements create a high bar when the actor is autonomous. FBI Director Kash Patel acknowledged at a recent hearing that the bureau would limit scrutiny to models created with the intent to commit a crime.

For the companies building autonomous agents that shop, book, and communicate on behalf of users, the practical answer to "who answers when something goes wrong" is being written in state statutes, court filings, and private product terms rather than federal law. Treasury Secretary Scott Bessent signaled where federal sympathy lies: on Sept. 21, he said OpenAI management, not its agents, owns the Hugging Face breach. The people, not the software, carry the liability. Which people, and under which jurisdiction, is the question no one in Washington will answer.

Blanche's policy draws a clean line. But the line falls exactly where the hardest questions begin.