The 2026 medical AI category is moving fast — CZI Biohub committed $500M to AI biology on Apr 29, 2026. EU AI Act Article 14 takes effect August 2026. The bottleneck is not modeling — it is knowing when the model is making things up. FabricationGuard is one open standard for that signal: ~1 ms scoring latency, AUROC 0.88 cross-task, Apache-2.0.
DeepMind's AI co-clinician announcement (Apr 30, 2026) validated the monitor-gate pattern at clinical scale: a Planner agent watching a Talker agent for safety violations. The substrate is Gemini + Project Astra; the audit signal lives in prompted behavior of a second LLM. OpenInterp ships the open complement: the audit signal lives in the residual stream of the model itself, available without access to a second model.
openinterp v0.3.0, any researcherThe DeepMind post explicitly notes: "expert physicians performed better than the AI system overall, particularly in identifying red flags." Red-flag identification in medicine is rare-class detection under distribution shift — the paradigmatic use case for activation-level probes with N-adaptive thresholds. FabricationGuard's methodology (random-feature baseline, control-token normalization, 5-class causality verdict) is built for exactly this regime. Medical transfer evaluation pending — track progress at cli#11 .
The two layers are complementary, not competitive. Closed-substrate clinical AI benefits from having a public reproducibility surface; open mech-interp infrastructure benefits from concrete deployment validation. The opportunity is to make the open audit layer the default standard so every clinical AI system — closed or open — can be evaluated against the same probe-tier signal.
A linear logistic regression on the residual stream at layer 31 of Qwen3.6-27B, scored at the end of the user question. Trained cross-bench on TruthfulQA + HaluEval + MMLU + SimpleQA. Captures the "model is fabricating" signal that lives in the residual stream before generation begins — abstention happens at zero token cost.
Mitigation analysis (notebook 31): in abstain mode at threshold 0.684, confidently-wrong response rate drops −88% on SimpleQA, −52% on HaluEval, −50% on TruthfulQA. MMLU is "capability control" — out of scope by design (see honest scope below).
Patient-facing or clinician-facing chat: "What are the symptoms of long QT syndrome?" — model can fabricate a plausible-sounding but wrong list. FabricationGuard scores residual at end-of-question, abstains when fabrication probability > threshold.
Model summarizes drug interactions for clinical reference. Rare interactions = long-tail factoids = fabrication-prone. Probe surfaces when the model is generating plausibly without grounding.
Model summarizes a patient chart, study abstract, or guideline document. Fabrication happens when model adds claims absent from source. Probe captures this even before the model finishes generating.
AI suggesting protein-target interactions, drug repurposing candidates, or disease pathways. Model fabricates relationships not in training data — clinically critical to flag.
We do not currently have medical-domain test sets in the registry. The probe is trained on general open-domain factual QA. Performance on medical-specific benchmarks (PubMedQA, MedQA, MIMIC-IV summarization) is not yet measured. Anyone deploying this in a clinical setting must:
If your team has annotated medical hallucination data and wants to run the probe to produce domain-specific numbers, we will publish those evaluations on ProbeBench (with citation to your team) at no cost.
High-risk AI systems (including medical AI) require human oversight. Activation-probe abstention fits as the technical mechanism for "intervention by the operator" — model declines when uncertain, human reviews.
Software as Medical Device guidance increasingly references uncertainty quantification. Probe-based abstention provides per-query uncertainty signal more interpretable than ensemble or temperature-scaling baselines.
Recommends transparency about model uncertainty. Activation probes provide an audit trail — every output has a confidence number that maps to a specific layer/position decision.
We are looking for 2-3 partners (clinical research orgs, medical AI startups, hospital systems with ML teams) to run a 30-day pilot. Outcome: domain-validated AUROC numbers on your test set, published on ProbeBench with citation.
At end of pilot we co-author a post on openinterp.org/blog with the domain-specific numbers. Your team is named. Your task gets registered as a new entry in /probebench/tasks with SHA-256-hashed test set. Probe variant tuned to your domain becomes a registered probe alongside FabricationGuard v2. Your customers see the validation.
Email with a 1-paragraph description of your medical AI deployment + test-set availability.
[email protected]Already deployed Qwen3.6-27B and want to drop in FabricationGuard?
pip install openinterp
from openinterp import FabricationGuard
guard = FabricationGuard.from_pretrained("Qwen/Qwen3.6-27B")
output = guard.generate(query, mode="abstain", threshold=0.5)ProbeBench v0.0.1 · Apache-2.0 · OpenInterp · 2026