“We’re prosecutors, not regulators” is more than a line; it’s the Justice Department’s attempt to draw a bright jurisdictional boundary in an era when artificial intelligence tempts every institution to become an AI agency. The core idea is straightforward: DOJ will chase crimes that involve AI, but it will not write the rules that govern AI itself.
The Short Version
- Attorney General Todd Blanche has articulated a prosecutorial—not regulatory—role for DOJ on AI: enforce criminal law, avoid writing technology rules.
- The position echoes DOJ’s earlier retreat from “regulation by prosecution” in digital assets, signaling institutional continuity rather than a one-off posture.
- DOJ still asserts an active lane: civil-rights enforcement, interagency coordination, and charging AI-related crimes when statutes fit.
- The live policy question is not whether DOJ governs AI, but how its charging choices and civil-rights authorities shape AI behavior without formal rulemaking.
What DOJ says its AI lane is—and isn’t
The Blanche doctrine, as publicly framed, rests on a clean division of labor. DOJ will investigate and prosecute violations of existing criminal laws where AI is part of the conduct, but it will not establish ex ante technical or market-wide rules for AI systems. The Attorney General has been explicit about the distinction—“We’re prosecutors, not regulators”—and has tied it to familiar criminal predicates: fraud, hacking, identity theft, and other offenses that do not require a bespoke AI code to charge. Reportage of Blanche’s remarks also asserts that DOJ has already brought AI-related cases, underscoring that the posture is not permissive; it is a venue choice about who writes rules and who enforces the penal code.
This approach is not invented for AI. Under Blanche, DOJ previously renounced “regulation by prosecution” in digital assets, signaling that criminal cases should not be used to superimpose regulatory frameworks better set by agencies with rulemaking charters. That policy—however one judges its merits—supplies institutional logic: use the criminal law for deceit, theft, and violence; leave technical guardrails and licensing to the regulators President Trump has tasked in other domains.
How the boundary fits within federal AI governance
Framed this way, DOJ’s role aligns with a long-running American compromise around fast-moving technology: Congress and sector regulators make the rules; prosecutors police violations of duly enacted statutes. The department’s own public AI materials, however, show that “not a regulator” does not mean “absent from governance.” DOJ inventories its internal AI use cases, commits to government-wide standards, and runs interagency convenings through the Civil Rights Division to coordinate how federal authorities address algorithmic discrimination. Those are governance acts—internal controls, public transparency, and civil-rights enforcement—without crossing into prescriptive technical rulemaking about models or training data across the private market.
That duality explains why critics talk past one another. Proponents of a stronger DOJ role point to civil-rights authorities and coordination mandates to argue the department already shapes AI behavior. They are right about the influence—and wrong that this turns DOJ into an AI regulator in the administrative-law sense. Blanche’s line draws on a principled constraint: prosecutors act where a statute attaches to conduct; regulators set ex ante standards that bind entire sectors. Those are different instruments, even when they converge on the same technology.
Mechanism: what “prosecute, don’t regulate” looks like in practice
In practical terms, the Blanche doctrine sorts AI matters into three buckets. First, ordinary crimes with novel tools—deepfake-enabled fraud, credential stuffing accelerated by model-assisted scripting, AI-aided IP theft—remain chargeable under existing statutes. DOJ’s message is: expect indictments here; no special AI rule is needed to pursue deception or intrusion. Second, civil-rights harms arising from automated systems trigger the Civil Rights Division’s portfolio: pattern-or-practice investigations, consent decrees, and interagency coordination to prevent unlawful discrimination in benefits or programs. That is enforcement of existing law, not promulgation of model standards. Third, market-wide safety, transparency, or model-governance mandates—what versioning to disclose, how to watermark outputs, which training datasets are permissible—fall to regulators and, ultimately, Congress. DOJ’s role in that third bucket is chiefly advisory and litigative, not legislative.
This partition also tempers a persistent policy temptation: using high-profile indictments to fill regulatory gaps. The digital-assets experience taught DOJ that aggressive theories at the edge of statutory text can backfire—muddying the law and chilling legitimate innovation. The current posture aims to avoid that trap while preserving the ability to act decisively when AI is the means of a classic crime.
Counter-arguments: doesn’t DOJ already “govern” AI?
There is credible evidence for a broader DOJ footprint. Department pages emphasize using AI to uphold the rule of law and protect civil rights, and the internal AI inventory reflects formal governance of the department’s own systems. In the prior administration, the Attorney General was also directed to coordinate enforcement of civil-rights laws related to AI and to review DOJ’s capacity to investigate rights deprivations tied to AI use. All of this signals an active DOJ, not a minimalist one.
None of that contradicts Blanche’s line. Coordination is not regulation; civil-rights enforcement is not model licensing; internal risk controls are not market-standard setting. If anything, these materials map the outer edge of DOJ’s legitimate authority in AI without collapsing into a de facto tech regulator. The harder unresolved questions—where civil-rights remedies end and technical mandates begin; how to draw causation when an algorithm mediates a harm—are where interagency process and courts, not unilateral DOJ edicts, should do the boundary drawing.
.@AGTODDBLANCHE ON AI: “WE’RE PROSECUTORS, NOT REGULATORS.”
Asked whether artificial intelligence needs greater regulation or even a slowdown, Todd Blanche says DOJ will investigate criminal violations involving AI, but regulating the technology is not the department’s job.… pic.twitter.com/BLSOgtr7ez
— LindellTV (@RealLindellTV) September 15, 2026
Implications for companies, developers, and the policy path ahead
For industry, the signal is crisp. Do not read “not a regulator” as a safe harbor. If AI is the instrumentality of fraud, market manipulation, identity theft, or obstruction, DOJ is promising to act. If algorithms drive disparate impact in federally funded programs, expect civil-rights scrutiny. What you should not expect is DOJ to prescribe technical standards for training data, model evaluation, or deployment controls—those demands, if they come, will arrive through sector regulators, NIST frameworks, or congressional statute, not a criminal charging memo.
For policymakers, the gap Blanche highlights is the one Congress must fill when it believes prevention requires ex ante obligations. Where ambiguity persists—synthetic media labeling, dual-use safeguards for frontier models, duty-of-care for high-risk deployments—clear legislative text and regulator-led rulemaking are the right tools. DOJ’s charging decisions will still shape behavior at the margins; prosecutors always do. But the department’s stated posture—anchoring AI oversight in existing criminal and civil-rights law—respects the separation of roles that keeps innovation policy from being made at arraignment.
Sources:
ground.news, justthenews.com, bloomberg.com, legalaiinsights.com, pbs.org, justice.gov, cbsnews.com, congress.gov