PRODUCT UPDATES · JAN 2, 2026 · 3 MIN READ
AI grading audit trail: every CTCAE grade now records the model that produced it
First update of 2026: every CTCAE grade records the language model that produced it, each graded event carries an organ-system category, and sign-in adds email and password.
Burna AI Team
What shipped in Week 1, 2026
Between December 29, 2025 and January 4, 2026, every CTCAE grade Burna AI produced began recording the identifier of the language model that produced it, and every graded event began carrying an organ-system category. Those two fields strengthen the AI grading audit trail that research coordinators, investigators and sponsor reviewers read when they need to know where a grade came from. The same week added email and password sign-in and brought the grading workflow to the provider mobile app.
What an AI grading audit trail needs to record
An audit trail for AI-assisted grading answers two questions about every grade: what produced it, and on what basis.
The first question has a regulatory baseline. Among the controls for closed systems, 21 CFR 11.10(e) lists secure, computer-generated, time-stamped audit trails that independently record the date and time of operator entries and actions that create, modify or delete electronic records. How 21 CFR Part 11 applies to AI-assisted adverse event grading covers the rest of the rule.
The second question is newer. FDA's January 2025 draft guidance, Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products, proposes a risk-based framework for establishing the credibility of an AI model for a particular context of use. Judging a model's credibility starts with knowing which model produced the output in front of you.
After this release, the grading output for every event carried these fields (what the audit trail includes lists the full set today):
| Field | The question it answers |
|---|---|
| Model identifier | Which language model produced this grade |
| Workflow version | Which version of the grading workflow ran |
| Rationale and evidence | Which findings in the note support the grade |
| Confidence score | How strongly the model held the grade, from 0 to 100 |
| Organ-system category | Which body system the event belongs to |
The model identifier on every CTCAE grade
From December 31, 2025, the grading workflow writes the identifier of the model it called into its output, and each grade's provenance block carries that identifier beside the workflow version. The same change made evidence, confidence and category required on every graded event, so each one arrives with all three beside its rationale.
Models change over the life of a trial. With the identifier on the grade itself, a quality reviewer can separate the grades produced before a model update from those produced after it, and a sponsor can ask which grades a given model produced. Both answers come from the grade, with no side log to reconcile.
Organ-system categories for every graded event
Each graded event also carries one of eleven organ-system categories: hematologic, neurologic, dermatologic, gastrointestinal, cardiovascular, pulmonary, renal, hepatic, metabolic, infectious and other. The category is assigned during grading, with no separate step for the coordinator.
It is a coarser grouping than the 26 MedDRA system organ classes that organize CTCAE itself, and it is built for the body-system questions a safety review asks, such as every cardiovascular event on a regimen this quarter. Anemia falls under hematologic, and Peripheral sensory neuropathy under neurologic.
Email and password sign-in
On December 30, 2025, the shared sign-in layer behind the web and mobile apps gained email and password sign-up, sign-in and password reset. It sits alongside the one-time email code already in place, so each team can use whichever its site prefers.
The grading workflow on the provider mobile app
On December 29, 2025, the provider mobile app gained the CTCAE workflow screens: enter a clinical note, follow the grading run step by step, and read the graded events and the drug attribution results when it finishes. Visits gained their own list and detail screens.
The same release, and the sign-in change the next day, moved duplicated interface code out of both mobile apps, the provider app and the patient app, and into shared packages, so a fix made once reaches both.
Since then
- The model identifier still travels with every grading run. When the model's category is missing or falls outside the eleven, the category now comes from the CTCAE system organ class of the matched term.
- A grade becomes a record when a qualified investigator signs it. The 21 CFR Part 11 electronic signature carries what was accepted, what was overridden and the citation set as signed. What Burna AI does today is the dated record of each capability.
- The provider mobile app is now CTCAE by Burna AI, redesigned on September 14, 2026. Trial participants now report through Healthnotes, the participant app sponsors deploy through their sites.
- The controls a security review asks about, Part 11 among them, are on the architecture page.
See it on your own workflow
To see how a grade, its model and its evidence travel together through review and signature, book a 15-minute conversation.