“Trust, but verify”: AI governance for IVF labs in India, from validation to vendor questions
15 November 2026 · IVF Automation · 6 min read
AI governance for IVF labs means treating every AI tool as a change you must prove, not a feature you switch on. Test it on your own cases before it touches a decision, keep a named person accountable for every output, log what the tool said and which version said it, and get the vendor's regulatory and data-protection position in writing.
That is roughly what “trust, but verify” looks like on a Monday morning in an Indian lab. The phrase is also the title of a talk at FERTIVISION 2026 in Kolkata, and it is a good test for every AI pitch you will hear there. Here is how to put it into practice.
What is the “Trust, but verify” talk at FERTIVISION 2026?
It is a faculty talk titled “Trust, But Verify”: Bringing Cold War Wisdom to AI in IVF, listed for Prof Sezcan Mumusoglu. FERTIVISION 2026 is the Indian Fertility Society's 22nd annual conference, on 18–20 December 2026 at Biswa Bangla Mela Prangan, Kolkata, with the theme “Practice with Precision”. The official site also lists Dr Siladitya Bhattacharya on whether AI and real-world data can revolutionise evidence-based practice. The day and time of the talk were not shown on the homepage when we checked on 15 November 2026. Other Indian and global meetings are on our IVF events calendar.
Why does AI in the IVF lab need verifying, not just trusting?
Because the evidence is still early, and models do not always travel well between clinics. The ASRM Practice Committee's 2026 opinion, AI in the IVF laboratory, says implementation is at an early stage and recommends proceeding with caution. It notes that the largest randomised trial so far could not show that a deep-learning model was non-inferior to standard morphology for clinical pregnancy, although it showed no harm and cut blastocyst assessment time almost tenfold. ASRM also points out that algorithms built on one set of clinics and patients may not give consistent results elsewhere, and warns about commercial pressure to adopt unproven AI. Our guide to AI in the IVF lab covers the trial evidence in more detail.
How should an IVF lab validate an AI tool before using it?
Run it in shadow mode on your own cases before it informs any real decision. ICMR's AI ethics guidelines say healthcare AI must undergo rigorous clinical and field validation before it is applied to patients, and that differences in training data can make different tools disagree. ASRM describes validation on new data followed by real-world testing, preferably a randomised trial. A practical local sequence looks like this:
- Write down the exact task and the vendor's stated intended use.
- Record a baseline from your own KPIs before go-live.
- Run the tool alongside your normal process for a set period and log where it agrees and disagrees with your team.
- Have the lab director review the results and sign off in writing.
- Repeat a shorter check after every software or model update.
Who stays accountable when AI is used?
A named embryologist or clinician, every time. ICMR says humans should have complete control of AI-based healthcare systems and medical decision-making, supports a “human in the loop” model, and says human oversight and manual override must be compulsory for AI capable of causing severe injury. It also says the use of AI should be disclosed to the person receiving care. In practice, your record should show what the tool suggested, who accepted or overrode it, and why.
Is AI software a medical device in India?
It can be, depending on its intended use. CDSCO's guidance on medical device software (CDSCO/MD/GD/MDSW/01/2026) says AI/ML-based software that meets the medical device definition is treated as medical device software under the Medical Devices Rules 2017, and that definition includes “control of conception”. The guidance expects the software label to show a version or build number and licence details. It also says performance can change over time with new patient populations, updates or model changes, so manufacturers must keep assuring performance across the product's life. We compare this with the FDA route in our AI embryo assessment FDA and CDSCO guide.
What do the ART Act and DPDP rules mean for AI records and data?
Keep AI outputs in your clinical records, and hold vendors to the same security standard you are held to. Section 23 of the ART Act 2021 requires clinics and banks to keep accurate records, retained for ten years. The DPDP Rules 2025 were notified on 13 November 2025 with an 18-month phased timeline, so most operative duties apply from 13 May 2027. Rule 6 covers processing done on your behalf by a data processor. It asks for access controls, logs and monitoring, backups, one-year retention of those logs, and security clauses in processor contracts. Our free DPDPA readiness check is a quick place to start.
What should you ask an AI vendor before signing?
Ask for evidence and paperwork, not slides.
| Question | What a good answer includes |
|---|---|
| What is the intended use, in writing? | Task, embryo stage or workflow step, compatible hardware |
| What is the CDSCO status? | Licence or registration number and risk class, or a clear statement that it is not a device and why |
| What evidence supports it? | Published studies, ideally prospective; populations similar to yours |
| How are updates handled? | Version numbers, release notes, advance notice of model changes |
| What is logged? | Each output with timestamp, version and user, exportable to your records |
| Where does patient data go? | Storage location, access controls, breach process, DPDP security clauses in the contract |
Frequently asked questions
How should an IVF lab validate an AI tool before using it?
Run it in parallel with your usual process on your own cases first, compare its output with your embryologists' decisions and your KPIs, and only then let it inform real decisions. ICMR says healthcare AI must undergo rigorous clinical and field validation before it is applied to patients.
Is AI software used in an IVF lab a medical device in India?
It can be. CDSCO's 2026 guidance says AI/ML-based software that meets the medical device definition is medical device software under the Medical Devices Rules 2017, and that definition includes control of conception. Ask the vendor for its regulatory status and licence.
Who is responsible when an AI tool gets it wrong?
Your clinic. ICMR's guidelines say humans should keep complete control of medical decision-making, so a named embryologist or clinician should sign off every AI-assisted decision.
Does the DPDP Act apply to AI tools that use patient data?
Yes, if they process digital personal data. Most duties under the DPDP Rules 2025, including security safeguards and one-year log retention, apply from 13 May 2027, and they extend to vendors processing data on your behalf through your contract.
Sources
- Indian Fertility Society. FERTIVISION 2026, 22nd annual conference, Kolkata, 18–20 December 2026 (faculty list: “Trust, But Verify”: Bringing Cold War Wisdom to AI in IVF). Accessed 15 November 2026.
- ASRM Practice Committee. Artificial intelligence in the in vitro fertilization laboratory: a committee opinion. Fertil Steril 2026;126:494–8.
- ICMR. Ethical Guidelines for Application of Artificial Intelligence in Biomedical Research and Healthcare, 2023.
- CDSCO. Guidance Document on Medical Device Software under MDR-2017 (CDSCO/MD/GD/MDSW/01/2026).
- Press Information Bureau. Government notifies the Digital Personal Data Protection Rules, 2025 (Rules G.S.R. 846(E), notified 13 November 2025; rule 6 on security safeguards).
- Government of India. The Assisted Reproductive Technology (Regulation) Act, 2021, section 23.
Your next step
Build AI, LIMS or lab software for IVF clinics in India? Claim a free vendor listing so lab managers can find you, and your compliance answers, when they shortlist.
General operational information only; not medical or legal advice and no claim about clinical outcomes. Session details are as published on the FERTIVISION 2026 website on 15 November 2026 and may change. See our disclaimer.