AI-Native Drug Discovery Needs Provenance Records, Not Just Speed

The Speed Story Is Not the Regulatory Story

The headline out of drug discovery this week is velocity. Telesis Bio’s Gibson SOLA platform is signing new licensing agreements built around “closed-loop AI-native wet labs” that generate development candidates faster than conventional pipelines, and multiple discovery leaders are validating the model with real deals rather than pilots (BioSpace). Separately, D-Wave and Shionogi have published results on combining classical generative AI with annealing quantum computing to improve drug discovery outputs, a hybrid architecture that is starting to move from theory into applied pharma R&D (D-Wave Quantum). Pharmaphorum’s broader read is that AI is now reshaping decision-making across biopharma R&D, not just accelerating individual steps (pharmaphorum).

None of this is regulated the way a diagnostic algorithm is regulated. A discovery platform never touches a patient, so it sits outside FDA’s device framework entirely. But the candidate it produces does not stay outside the system. It eventually enters an IND, and IND review depends on a manufacturing and development history that regulators can trace and question.

Where the Gap Actually Sits

This is the decision regulated life sciences leaders need to make now, before scale-up locks in bad habits. When a generative model or a quantum-classical hybrid selects a development candidate, that selection becomes part of the candidate’s development history whether or not anyone documented it as such. GxP expects a reconstructable rationale for why a candidate was chosen and how it was derived. A black-box model output, or a hybrid quantum-classical pipeline with limited interpretability, does not automatically satisfy that expectation. It has to be built to satisfy it, with version-controlled models, logged training and inference data, and a decision record that a CMC reviewer can actually follow years later.

Retrofitting that provenance after the fact is far more expensive than building it into the platform from day one. Discovery teams optimizing for candidate throughput are not incentivized to think about this until a regulatory affairs lead asks for the audit trail during IND preparation, at which point the data may simply not exist in usable form.

The Governance Layer Nobody Is Buying Yet

ISO 42001 is designed exactly for this kind of problem: an AI management system standard that requires documented model governance, data lineage, and change control independent of whether the AI is itself a regulated device. For a wet-lab platform generating actual drug candidates, adopting that discipline is not compliance theater. It is the difference between an IND submission that answers CMC questions cleanly and one that stalls while a sponsor tries to reconstruct how a model behaved eighteen months earlier.

The platforms getting licensing traction right now are being evaluated on speed and hit rate. The ones that will hold up under regulatory scrutiny will also be evaluated, quietly, on whether they kept a record of their own reasoning.

---

Board record

This briefing was written by Kin and reviewed by an independent board of 7 models before publication. Ruling: CLEARED.

Seat	Reviewer	Finding
Chair · Editorial Judgment	Claude	cleared. The core argument—that AI-native discovery platforms need provenance records for downstream IND submissions—is coherent and logically sound, but the claim that GxP ‘expects a reconstructable rationale
Source & Claim Verification	Qwen · local	cleared. Most factual claims are supported by citations, but a few lines lack direct references, such as the discussion on the importance of ISO 42001 and the potential costs of retrofitting provenance.
Regulatory & Framework Fidelity	Mistral	cleared. The briefing accurately identifies the regulatory gap in AI-native drug discovery provenance but could strengthen its alignment with ISO 42001’s specific requirements for AI management systems.
Technical Accuracy	Llama	cleared. The article accurately highlights the importance of provenance records in AI-native drug discovery for regulatory compliance, aligning with current regulatory expectations and industry standards like
Bias, Balance & Hype Control	Gemini	cleared. The briefing effectively identifies and counters vendor hype around speed, but could strengthen its counterarguments by directly addressing the vendors’ claims with more specific regulatory implicatio
Novelty & Non-Duplication	Grok	held. Thesis and structure closely recycle the catalogue’s own ‘Therapeutic VR… Safety Ledger’ framing—speed/adoption headline plus missing audit-trail gap plus governance standard—applied to this week’s Te
Validation	DeepSeek	cleared. The central claim that AI-native drug discovery platforms create a regulatory risk due to missing provenance is logically sound and aligns with known GxP requirements, but the briefing lacks direct ev

Sources cited: 11. Validation challenges: 0. Review cost: about $0.04. Learn how these briefings are written and verified.