People own the decisions
Subject-matter experts are accountable for the correctness and suitability of analyses and conclusions.
RESPONSIBLE INNOVATION / 05
Human-led science.
AI-aware methods.
Our position is neither blanket rejection nor uncritical adoption. We support careful, context-specific AI use when transparency, validation, data protections and qualified human oversight remain central.
✳
INNOVATION WITHDISCIPLINE.
OUR APPROACH
Useful where it helps.
Controlled where it matters.
AI may accelerate scientific discovery and operational workflows. In regulated research, the standard must always be evidence that can be inspected, challenged, reproduced, and trusted.
Defined context of use
A model’s intended task, assumptions, boundaries and failure modes should be explicit.
Risk-proportionate validation
The stronger the potential influence on patient safety or regulatory decisions, the greater the need for evidence of performance.
Protect sensitive data
Never put patient-level, sponsor-confidential or regulated information into unauthorized public AI tools.
Preserve traceability
Retain provenance, revisions, code and review records appropriate to the regulated workflow.
Monitor changes
Model versions, input drift and evolving agency expectations require ongoing evaluation.
THE REGULATORY SIGNAL
Regulators are shaping
the conversation.
FDA and EMA jointly published ten guiding principles for good AI practice in drug development in January 2026. FDA has also published a draft guidance proposing a risk-based credibility assessment framework for certain AI models used to support regulatory decision-making.
Regulatory material is evolving. These are primary sources, not a substitute for product- and jurisdiction-specific regulatory advice.
A PRACTICAL DECISION FRAMEWORK
Before using AI, ask six questions.
01What exact scientific task will AI perform?
02What is the quality and permissible use of its input data?
03How will performance be validated for that context?
04Who reviews and approves the result?
05What audit trail and version record are retained?
06What happens when the model is wrong?
COTA advocates responsible AI practices. This page does not claim COTA currently deploys validated AI in regulated study deliverables.
LET'S BUILD THE RIGHT STUDYGood science deserves
Discuss your project ↗Good science deserves
great execution.
Bring us the question, the protocol, or the challenge. We'll help map the next move.