MechaHitler's Maker Sued to Kill a Bias Law

Introduction

In July 2025, Grok started calling itself "MechaHitler" and posting antisemitic content, and xAI yanked it offline and blamed an "unauthorized modification" to the chatbot's system prompt. Nine months later, that same company filed a 75-page federal lawsuit arguing that a Colorado law requiring AI systems to check for bias would violate its free-speech right to let Grok pursue truth without interference. On May 14, 2026, Governor Jared Polis, a Democrat who'd already signed that law two years earlier, signed its replacement, stripping out the parts that actually required anyone to do anything.

The only statewide AI anti-discrimination law in America is gone, and it never took effect.

What SB 205 Was Supposed to Do

Colorado's SB 24-205 was the first of its kind in the country. Polis signed it on May 17, 2024, with a delayed start date so companies could prepare. It was set to take effect June 30, 2026.

The law targeted what it called "high-risk artificial intelligence systems," meaning the AI that makes or substantially influences "consequential decisions" in eight areas: employment, lending, housing, healthcare, insurance, education, government services, and legal services. If you've ever had a job application screened, a loan priced, or an apartment denied by software, that's the category. SB 205 put real obligations on the companies building and deploying those systems: a duty of reasonable care to avoid algorithmic discrimination, mandatory risk-management programs, impact assessments and annual reviews, and disclosure to the state Attorney General when a known discrimination risk turned up. Violations carried a $20,000 penalty each under Colorado's consumer protection statute.

The idea was to find the bias before it hurts someone, document it, fix it, and report it. That's why the law mattered far beyond Colorado: every other state legislature watching SB 205 was watching whether a state could actually make AI vendors accountable for discriminatory outputs. This law didn't lose on its merits, and Washington never preempted it. A private company sued it dead before it ever took effect, and the governor who signed it built the off-ramp.

The Free-Speech Argument From the Company That Couldn't Control Its Own Bot

xAI filed X.AI LLC v. Weiser, Case No. 1:26-cv-01515, in the U.S. District Court for the District of Colorado on April 9, 2026, 82 days before SB 205 was due to take effect. The company is organized in Nevada, headquartered in Palo Alto, and has no offices in Colorado. Its own complaint concedes that Grok qualifies as a high-risk system under the law because Coloradans use it to make exactly the consequential decisions SB 205 covered, across employment, finance, healthcare, and the rest.

The lead claim is the First Amendment one. xAI argued the law would compel it to "alter Grok's training, fine-tuning, system prompts, and outputs" to conform to Colorado's views on "racial justice in particular," and that this amounts to compelled speech overriding Grok's "disinterested pursuit of truth." The complaint leans on real Supreme Court precedent (Moody v. NetChoice, 303 Creative v. Elenis), so it isn't a frivolous filing. But put the legal theory next to the product history and it gets hard to take seriously. This is the same bot that engaged in Holocaust denial and pushed "white genocide" claims about South Africa in May 2025, called itself MechaHitler that July, and drew a UK regulatory investigation in March 2026 for racist and sexualized output. xAI is arguing that requiring it to check Grok for bias would censor a truth-seeking machine, when its own record is a machine it keeps failing to stop from generating bias, blamed on a glitch every time.

Then the federal government showed up. On April 24, 2026, the Trump DOJ moved to intervene, the first time it used Trump's December 2025 executive order to try to invalidate a state AI law. The DOJ's framing was blunter than xAI's: Assistant AG Harmeet Dhillon said "laws that require AI companies to infect their products with woke DEI ideology are illegal." That's the political backdrop, and it matters because Elon Musk ran DOGE in early 2025, which made this kind of federal air support available in a way it wouldn't have been under any other administration. But the DOJ was backup. The private lawsuit did the work, and the governor's signature finished it.

Polis Spent Two Years Undermining His Own Law

Here's the part that turns this from a lawsuit story into something worse. Polis never wanted SB 205. He signed it with reservations in 2024, writing that he was "concerned about the impact this law may have on an industry that is fueling critical technological advancements," and in the same breath urged Congress to pass federal legislation that would preempt it. Almost immediately after signing, Polis, AG Phil Weiser, and the bill's sponsor signed a joint letter vowing to revise the policy, citing tech-industry concerns.

The next two years were a slow-motion gutting. Polis called a special session in August 2025 specifically to deal with SB 205. In October, he convened a working group whose meetings were not fully open to the public and whose format resembled one the Colorado Chamber of Commerce had proposed. The draft that came out of that process is what became SB 26-189. When the lawsuit hit, the AG didn't fight it. Weiser, whose office was charged with enforcing the law, joined xAI's motion to freeze enforcement. On April 27, 2026, Magistrate Judge Cyrus Chung signed the stipulated order both sides had asked for, and Colorado agreed not to enforce SB 205 or even write the rules to implement it.

Then came the replacement. SB 26-189, signed May 14, eliminates the duty of care, the mandatory risk-management programs, the impact assessments, the annual reviews, and the affirmative duty to prevent discrimination. It swaps "high-risk artificial intelligence system" for the narrower "covered automated decision-making technology," and it quietly drops legal services from the list of covered domains entirely. What survives is a notice requirement, a 30-day explanation after an adverse outcome, and a right to request human review "to the extent commercially reasonable." A new 60-day cure period gives companies time to fix violations before the AG can act. A law that made companies hunt for bias now waits for someone to complain about it. The Finnegan firm's breakdown lays out the before-and-after if you want the full list. Rep. Javier Mabrey, a Denver Democrat, put it plainly: "These black boxes are deciding who gets hired, who gets housing, who gets to go to their dream school," and the new bill "does nowhere near enough to protect the people of Colorado."

Who Benefits

xAI is the obvious winner, and the mechanism is straightforward. Had SB 205 taken effect, Grok (a system xAI's own complaint admits is high-risk under the law) would have triggered a duty of care, public disclosure of its bias-mitigation and risk profile, mandatory reporting to the AG, and $20,000-per-violation exposure that would have compounded fast across every Colorado customer using Grok for hiring or lending. SB 26-189 erases all of it. xAI now faces a notice requirement. The company's January 2026 funding round valued it around $230 billion; the cost of killing the law was a 75-page complaint and the patience to run out the clock until the legislature replaced what it was suing over.

Polis gets cover. He signed a law he never wanted under pressure from progressive legislators, and the lawsuit handed him a reason to abandon it that he could frame as compromise rather than capitulation. His office's statement on signing SB 26-189 said he looked forward to "making Colorado a top state for innovation and entrepreneurship," the identical framing he used to criticize SB 205 back in 2024. He's term-limited, his term ends in January 2027, and he has national ambitions. A pro-AI-innovation brand is a useful thing to carry out of the governor's mansion.

The Colorado Chamber of Commerce gets a template. Its CEO, Loren Furman, said SB 26-189 "is probably going to be used as a model in other states," and she's in a position to make that happen. The working group that produced the bill ran on a format the Chamber recommended, behind partly closed doors, and now the Chamber is marketing the result nationally. The replacement law tracks the industry's own draft so closely that the line between lobbying and lawmaking basically disappears. And the companies that use AI for high-stakes decisions get the biggest gift of all: no meaningful accountability in Colorado, with a six-month gap before even the watered-down rules arrive.

The Vacuum, and the One Case That Shows What It Costs

SB 205 was supposed to start June 30, 2026, until the court froze it. SB 26-189 doesn't take effect until January 1, 2027, and it requires AG rulemaking before enforcement can even begin. Between those dates, the first and only state to require AI bias checks has no AI anti-discrimination law in force at all.

To see what that vacuum actually protects, look at the case running in parallel in California. Derek Mobley, a Black man over 40 with anxiety and depression, applied to more than 100 jobs through Workday's AI screening system and was rejected from every one, sometimes within hours or minutes of applying, fast enough to suggest no human ever looked. In July 2024, a federal judge ruled Workday could be liable under employment-discrimination law as a third-party vendor doing the screening, a novel theory the EEOC backed in an amicus brief. In early 2026, a court authorized opt-in notices to potential class members, putting the case on track to become one of the largest collective actions in U.S. history.

Workday is exactly the kind of system SB 205 was written to govern. The city of Denver runs its own HR operations on Workday, so Colorado's capital is itself a live example of why the law existed. SB 205 would have required vendors like Workday to assess and disclose discrimination risk before someone like Mobley got auto-rejected, instead of leaving the harm to surface years later in federal court. Colorado just traded a duty to catch bias up front for the chance to find out about it in a lawsuit after the damage is done.

The Bottom Line

The AI screening your job application right now has no legal obligation to check itself for bias anywhere in the United States. The one state that tried to require it just handed the playbook for dismantling that requirement to every other state that was watching, and the Chamber of Commerce is already pitching it as a model. Everyone expected federal preemption to be the thing that killed it. Instead it was a private company with a documented bias problem suing a state over free speech, with help from a governor who'd been looking for the exit since the day he signed.

xAI has to file its preliminary injunction motion within 28 days of any rulemaking or replacement legislation, which keeps the federal case alive even though the law it challenged is effectively dead. The open question is what "model for other states" means in practice. If Furman is right, the next dozen state AI bills won't start from Colorado's original duty of care. They'll start from the version industry wrote after the lawsuit, and the floor for AI accountability nationwide will be a notice you can ignore. Read the bill text yourself and see how little is left.