The Real Autonomous Diagnostic Threshold Hasn't Been Crossed — Here's What Would Cross It

Let's establish the definition before the argument. IDx-DR demonstrates the FDA can and will authorize autonomous AI diagnostics. It is not, however, what our forecast is tracking. Our 56% reflects the probability of autonomous AI diagnostic deployment reaching a new threshold along two specific dimensions: first, deployment of fully autonomous diagnostics across multiple high-acuity clinical domains (beyond the single-specialty, ophthalmology-specific model IDx-DR represents); second, clearance under a regulatory pathway that explicitly removes mandatory physician sign-off from the workflow loop for high-acuity diagnostic conclusions — not just image interpretation in a narrow screening context. Two other devices frequently cited in this space also fall short. Viz.ai's stroke triage clearance is explicitly assistive — it flags studies for radiologist review, not a diagnostic conclusion. Aidoc's foundation-model clearance, which we previously cited as 'a qualitative step toward autonomous diagnostic reasoning,' requires the same correction: Aidoc is cleared for workflow prioritization, not diagnostic conclusion. That framing was imprecise. These devices are proximate evidence that the FDA is building institutional fluency with AI submissions — a necessary but not sufficient condition for crossing the autonomy threshold we're forecasting.

With the definition locked, what does today's news actually tell us? Google DeepSomatic's work at Children's Mercy Hospital — confirming known leukemia variants while detecting 10 previously missed mutations — is the strongest piece of evidence in today's cycle for our thesis, but it's proximate evidence, not direct. It demonstrates AI diagnostic capability in a high-acuity oncology setting that no currently cleared device approaches. What it doesn't demonstrate is a regulatory submission for autonomous deployment, nor any signal from FDA's Center for Devices and Radiological Health that such a submission is imminent. DeepSomatic is publicly available as a research tool; it is not in a pre-submission meeting for autonomous clinical deployment. The Generate Biomedicines and Chai Discovery pharma deals are even more distal — they demonstrate AI's role in drug discovery workflows, which is a different regulatory category entirely. We weight today's evidence as confirming the technical capability trajectory but not the regulatory pathway trajectory.

Now for the counterargument we take seriously. The AMA's opposition to removing physician oversight isn't merely professional protectionism — it's backstopped by a malpractice liability framework that has no clear mechanism for assigning fault when an autonomous AI diagnostic produces a conclusion that harms a patient with no supervising physician in the loop. This is arguably a harder bottleneck than AMA policy preference, because the FDA has historically been reluctant to authorize autonomous devices where post-market liability is structurally unresolved. We are not aware of active federal legislation creating an AI diagnostic liability framework. At the state level, the 12-state wave of AI personhood exclusion laws we've tracked in prior cycles signals legislative bandwidth oriented toward containment, not liability clarification. This is the part of our thesis that keeps us up at night: the 443-day runway is generous for technical and regulatory submission timelines, but it may not be enough time for liability law to evolve sufficiently to give FDA political cover to authorize a multi-domain autonomous diagnostic.

What drives the 56% rather than, say, 40% or 70%? We should be honest that there's no clean base rate for this reference class. The FDA has created genuinely new autonomous device categories infrequently — IDx-DR itself took roughly a decade from research to clearance. If we anchor to the pace of novel regulatory category creation, the number might argue for something closer to 40%. We're above that base because: the 443-day runway is substantial relative to historical AI clearance timelines once a submission is filed; the FDA's CDRH has publicly engaged with the question of autonomous device pathways in draft guidance work since 2021; and the technical capability evidence (DeepSomatic, pharma AI deals, the sheer diagnostic AI investment surge visible in today's news) creates a level of industry pressure that did not exist when IDx-DR was being evaluated. But 56% is closer to a genuine uncertainty band than a confident prediction. What would move us to 65%: formal FDA draft guidance explicitly defining an autonomous diagnostic pathway without mandatory physician sign-off, or a high-profile CDRH public statement that the agency is prepared to evaluate multi-domain autonomous submissions under an existing pathway. What would drop us below 45%: Congressional action codifying physician-in-the-loop requirements for FDA-cleared diagnostics, or a high-profile autonomous AI diagnostic failure triggering an agency-wide review moratorium.