While Washington Debates the Robot Takeover, AI Is Already in Our Exam Rooms

If you caught the news this week, you saw Washington in full AI panic mode. Senators racing to introduce guardrail legislation before the midterms. A researcher resigning from a major AI company with dire warnings. Bills to ban superintelligence. The conversation is loud, urgent and focused almost entirely on one question: how do we keep AI from spinning out of human control someday?

We’re not here to dismiss that conversation. But we are here to tell you not to get so distracted by the science fiction scenario that you miss the one already playing out. Because while Washington debates what AI might do to humanity in the future, AI is making decisions about our health care right now. And nobody asked us.

Earlier this year, physician and Health Justice founder Dr. Oni Blackstock laid it out in STAT News, and her piece deserves more attention than it got. After two major AI companies launched dedicated health care initiatives in January, a study found that one of the new AI health tools had a 50 percent error rate, incorrectly recommending that care be delayed in emergency test cases half the time. Half. The. Time. And that error rate wasn’t identified before the tool was rolled out.

This isn’t new territory for us. A widely cited algorithm affecting an estimated 200 million Americans systematically underestimated how sick Black patients were because it used medical expenses as a measure of illness, and patients had no idea the tool was shaping their care. Think about that. A machine looked at the fact that our health care system historically spends less on Black patients and concluded we must be healthier. That’s not intelligence. That’s bias wearing a lab coat.

The consequences show up in trust, and trust is already in crisis. A national survey of more than 443,000 U.S. adults found trust in physicians and hospitals fell from 72 percent to 40 percent between 2020 and 2024. For Black communities, that collapse lands on top of generations of earned mistrust, from Tuskegee to Henrietta Lacks to the dismissed pain that too many of us have experienced in an exam room this year. And mistrust isn’t just a feeling. Patients who distrust their providers are more likely to delay care, including preventive screenings, and to stop taking their medications, patterns associated with higher rates of hospitalization and premature death. Now layer AI on top: 66 percent of Americans surveyed reported low trust in their health care system to use AI responsibly, and patients who have experienced discrimination in health care are significantly less likely to trust health systems to use AI responsibly.

So yes, Congress, pass the guardrails. But hear us clearly: guardrails that only protect humanity from a hypothetical future takeover, while leaving Black women unprotected from the algorithms denying our claims and downgrading our symptoms today, are guardrails with a hole in them. That’s why Technology Access and Artificial Intelligence is one of the five pillars of our National Health Policy Agenda. Equity can’t be a software update we install after the harm is done.

Dr. Blackstock’s prescriptions point the way, and we co-sign every one. Patients and community members need formal decision-making roles, not just advisory positions. Health systems and insurers need to publicly report how these tools perform across racial and ethnic groups before they’re rolled out. And patients need to be told clearly, and in advance, when AI is being used in their care.

AI could genuinely make health care better, and we want that future. But the companies building these tools have the capacity to move fast. The harder work is moving at the speed of trust. For Black women, trust isn’t given. It’s earned, receipts required. We’ll be watching, we’ll be at the table, and we’ll keep asking the question that matters most: who was in the data, and who wasn’t?