What happens to AI devices after clearance.
Adverse-event trends, recalls, and performance-drift signals, mapped to each device on FDA's AI-enabled device list. Every signal is an observable with a source — never a judgment.
Live cohort — snapshotted AI/ML devices
devices snapshotted — refreshed weekly, rolling
Latest snapshot 2026-07-08
Of devices where the trend is computable.
Own line re-cleared, own applicant's recall, or predicate neighbors.
Includes a software recall anywhere in the device's product code.
Of devices scanned while the letters feed was live.
Every figure is a presencerate with its denominator inline — never pooled across devices. Recall and MAUDE counts are product-code-level (MAUDE reports are not reliably attributable to a specific device). Descriptive only: “this code had N software-related recalls” — never “this device is unsafe.”
Drift signals by kind
Machine access
Query the same cohort over MCP — device_postmarket_lookup, postmarket_search, cohort_postmarket_stats
curl -s https://radar.healthai.com/api/mcp \
-H 'content-type: application/json' \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/call",
"params":{"name":"postmarket_search",
"arguments":{"panel":"Radiology","has_drift_signal":true}}}'The premarket half lives at Precedent — same corpus, same K-numbers, clearance to runtime.
Get drift-signal updates for the cohort
Weekly snapshots surface new drift signals — adverse-event inflections, re-clearances, software-recall patterns. Get notified when the cohort picture moves.