AI for drug safety

Your team confirms or corrects the draft, instead of building each case from scratch.

Watch one case move through the work
Vigil Safety
C.1.8.1 LIT-2026-00318 Literature item Clock starts once the ICSR is valid Owner SR
SRC Source review · click a finding to see it boxed on the page
J Immuno-Oncol · 2026 · case series

Immune-mediated hepatitis and colitis with anti–PD-1 therapy

We describe a 64-year-old man with NSCLC on an anti–PD-1 antibody who developed Grade 3 immune-mediated hepatitis and immune-mediated colitis twelve days after cycle 4. Both improved off-drug and on corticosteroids.

Patient 1 (pneumonitis, no descriptor) and Patient 3 (thyroiditis) are reported separately.

Valid-ICSR triage · one report per patient
P1Not validno descriptor
P2Validthe case we follow
P3Separatethyroiditis · C.1.10.r
Identifiable reporter Identifiable patient Suspect drug At least one reaction
Seriousness Serious · proposed E.i.3.2 a–f

Six criteria, at event level. Two apply; click a ticked one to see it on the source.

Results in death
Life-threatening
Caused / prolonged hospitalisation fax + call
Disabling / incapacitating
Congenital anomaly
Other medically important condition article
Possible duplicate match 0.9

Matches existing report US-PREMISE-2026-000492 on age, sex, country, drug, reactions, onset. Brand and batch are absent, so identity is not confirmed. Initials masked, not used as a key.

Draft · pending reviewerLink follow-upOpen new

How it works with your team

The agent drafts. Your team decides.

The agent is validated software that proposes. Your qualified reviewer decides, signs, and stays the author of record. It never writes to your safety system.

Draft quality is measured against a clinically labeled gold set, safety-critical errors held to the tightest tolerances, and reported before any efficiency claim. Target: up to ~60% lower cost per case, at held quality, proven on your data in a pilot.

Built for regulated drug safety

The questions your safety, quality, and security teams will ask.

Designed to support 21 CFR Part 11 EU Annex 11 ICH E2B(R3) GVP Module VI GAMP 5

Integration

  • Drafts E2B(R3) into your safety system (Argus, Veeva Vault Safety, ArisGlobal), yours or your provider's.
  • No new tool, no migration. The agent never writes to the system of record.
  • Your data and audit packets export any time.

Validation

  • An AI that drafts into a safety record is GAMP 5 Category 5, and validated that way.
  • One change-controlled config: model, prompts, MedDRA and WHODrug versions. Changes trigger re-qualification.
  • We supply the qualification evidence; you validate for your intended use.

Security

  • AWS and Amazon Bedrock in your tenant and region, encrypted in transit and at rest.
  • Reads are scoped, logged, least-privilege, or against an export you provide. No standing write access.
  • No training on your data. DPA, sub-processor list, SDEA, AWS BAA.

Audit trail

  • Append-only log of every action, each value linked to its source.
  • Model and config version stamped into each record's packet.
  • Part 11 e-signature: identity, UTC timestamp, meaning, tamper-evident binding.
  • The binding ICSR signature is applied in your system after import.

Who you'd work with

I'm Siddish, the founder. You'd work with me directly, on a pilot scoped to your acceptance criteria and your data.

Siddish, founder · hello@premise.bio

Put it on a case of yours.

The pack: a security overview, the validation plan, a DPA and sub-processor list, and a sample audit packet.