In April, the US Department of Justice did something extraordinary: it sided with Elon Musk's xAI against Colorado, jointly suing the state to block an AI anti-discrimination law. The federal government's intervention marks the first time it has directly challenged a state AI regulation in court.
Colorado's Senate Bill 205 was straightforward. It required companies to audit their AI systems for bias before deploying them in high-stakes decisions about hiring, housing, and healthcare. The business community pushed back hard. The state softened the bill in March, narrowing disclosure requirements before any lawsuit was even filed. It didn't matter.
The federal case rested on a curious argument: that checking whether AI discriminates against protected groups constitutes discrimination itself. The Justice Department called the law "state-mandated discrimination," claiming it "obligates AI developers and deployers to discriminate." The logic assumes AI systems operate on neutral criteria that shouldn't be altered. But the evidence says otherwise.
A 2019 study in Science examined a widely used healthcare algorithm running across US hospitals. The system assigned Black patients roughly half the care of equally sick white patients. The culprit: the algorithm used healthcare costs as a proxy for health needs, embedding existing racial inequities into clinical decisions. When researchers removed that flawed proxy, the bias disappeared and outcomes improved. The system didn't become less efficient. It got better.
Similar patterns have surfaced elsewhere. Models trained on less data from certain populations perform worse for those groups. This has been documented in facial recognition research and in studies of large language models. When companies increased representation in training data, performance gaps nearly vanished. None of this suggests that fixing bias breaks AI. It suggests that ignoring bias breaks trust.
xAI argued the Colorado bill would force Grok to promote the state's "ideological views on racial justice." Brianna Titone, the bill's lead sponsor in Colorado's legislature, offered a clearer framing: "SB 205 is about consequential decisions. We're not restricting speech. Our bill does not say that Grok still can't be a dick." She's right. Auditing for discrimination and improving outcomes has nothing to do with controlling what a chatbot says.
The other justification floating through Washington is speed. The Trump administration framed AI safety rules as obstacles to winning a global competition with China for artificial general intelligence. But a hiring algorithm that locks out qualified candidates, or a healthcare system that misallocates resources, isn't competitive. It's broken. There's no tension between making sure AI works fairly and making sure it works well.
On the business case against the law: Colorado's governor said far more companies are moving to the state than leaving. The Wall Street Journal found one example: Palantir cited the bill in a regulatory filing but never said it was abandoning Colorado because of the rule. Microsoft, by contrast, has flagged AI bias as a material business risk. The argument that anti-discrimination rules kill entrepreneurship lacks evidence.
The pressure campaign worked anyway. On May 14, Governor Jared Polis signed a replacement bill, SB 189, that repeals the original measure. Gone are requirements for companies to proactively assess high-risk AI systems for bias, conduct annual reviews, or report discrimination to the state. What remains: a requirement to share technical documentation with deployers (not the public) and a right for consumers to request human review, though most won't know that right exists. It's a retreat dressed as compromise.
The broader message is chilling. If the Justice Department can join a billionaire's lawsuit to kill a state law protecting consumers, other states considering AI safeguards will think twice. The federal government has signaled that transparency and accountability in high-risk AI aren't worth defending.
Author James Rodriguez: "This case reveals something darker than a policy disagreement: a coordinated effort to reframe basic consumer protection as ideological warfare, and a federal government willing to enforce that framing in court."
Comments