On October 1, 2026, Judge Alvin K. Hellerstein of the US District Court for the Southern District of New York denied the government's motion to dismiss UAW v. Department of State. According to the Electronic Frontier Foundation's account of the ruling, the court found that the threat of adverse immigration consequences is "certainly enough to 'deter a person of ordinary firmness from the exercise of First Amendment rights.'" It also found that a program alleged to silence members "strikes at the unions' representational function itself." This is a ruling on a motion to dismiss, not a decision on the merits. But the question it lets the case reach is one policymakers should take seriously: what happens to speech when a government says it is using AI to read it?
The case for the government's position
The government's argument deserves a fair statement. Consular officers decide who may enter and remain in the country, and Congress has long given the executive wide latitude over those decisions. Public social media posts are public. Reviewing them to screen for security threats is not obviously different from reading a published op-ed. The government has also maintained, as NBC News reported when the suit was filed on October 16, 2025, that noncitizens "do not have the same First Amendment rights as American citizens." A court should be wary of micromanaging national-security vetting, and no one disputes that a visa holder who incites violence or supports terrorism can be removed.
What the suit actually alleges
The supplemented amended complaint, filed May 14, 2026, does not challenge screening for threats. It targets what it calls the "Challenged Surveillance Program": an interagency State Department and DHS effort that, the plaintiffs allege, uses "automated technologies and artificial intelligence" to conduct "viewpoint-based surveillance" of visa holders and Lawful Permanent Residents. The complaint alleges that officials paired this with public threats to punish speech, followed by viewpoint-driven enforcement actions.
The plaintiffs are the UAW, the Communications Workers of America and the American Federation of Teachers. The complaint says the UAW alone has nearly 1,000,000 active and retired members, about 120,000 of them workers in higher education, a population with many noncitizens. It alleges that members have deleted posts, left leadership roles and skipped organizing events. The government has not conceded any of this, and these are allegations at the pleading stage.
The scale matters for the policy question. NBC reported that more than 55 million people hold valid US visas and are potentially subject to this kind of monitoring. A program that reaches that population does not need to punish many people to change how many of them speak.
Why the chilling-effect theory is serious
The government's likely response is that no one has been punished for the plaintiffs' speech, so there is no injury. Hellerstein's "person of ordinary firmness" standard rejects that framing. A chilling-effect claim does not require the government to act against every speaker. It requires that the threat be credible enough to change a reasonable person's behavior.
The Supreme Court has recognized this logic. In Americans for Prosperity Foundation v. Bonta (decided July 1, 2021), it held that California's donor-disclosure requirement was facially invalid because it burdened First Amendment rights and was not narrowly tailored to an important government interest. The Court looked at the burden on association and was not satisfied by the state's convenience. Bonta involved citizens and a disclosure rule, not immigrants and surveillance. But the principle carries over: when the government collects information about who says and associates with what, the cost is borne by speech that never gets uttered.
The claim also reaches beyond noncitizens. The complaint alleges that citizens are staying quiet too, for fear of exposing noncitizen relatives, friends and colleagues. If that is shown, it means the effect is wider than the immigration status of any one speaker.
The innovation angle
Our editorial position favors proportionate, evidence-based rules, and that is the standard this program should meet. Automated systems that flag "disfavored viewpoints" are a poor fit for a legal system built on identifying conduct. Language models and classifiers misread sarcasm, quotation, translation and political context. Applied to millions of accounts, even a low error rate produces many false flags, and each flag carries immigration consequences rather than a mild inconvenience. Nothing in the public record we reviewed shows how accurate these tools are or what review safeguards exist. That gap is itself a policy failure.
There is also a competitiveness cost. The US attracts researchers, engineers and graduate students in part because it protects open inquiry. A vetting regime that tells them to scrub their accounts and avoid political topics works against that. Universities and technology employers depend on those workers.
What to watch
The ruling does not decide whether the program is unconstitutional. The next stage will be discovery and merits briefing, where the government may have to describe what the tools do, what they flag, and how flags become enforcement decisions. That disclosure may matter as much as the eventual judgment, because the public currently knows little about how the system works.
A workable standard would keep security screening focused on specific threat indicators and conduct. It would require published accuracy testing and human review before any adverse action. And it would state in writing that lawful political expression is not a ground for removal. National security vetting can survive that standard. A program that depends on people not knowing how it works is harder to defend.