Celebrate Methods, Demand Decisions: What Nature Reviews Drug Discovery Gets Right About AI

Nature Reviews Drug Discovery's August 7, 2026 Perspective lands like a referee's whistle mid-match: the methods are flashy, the papers are plentiful, but the clinically relevant scoreboard is still mostly empty.

This opinion draws on the published Perspective by Bender, Thomas, Scannell, and colleagues. No private interviews were conducted.

The uncomfortable sentence

The authors write that despite a wide variety of AI methods developed, applied, and benchmarked, "evidence of their clinically relevant impact is, so far, disappointingly limited."

That is not anti-AI skepticism. It is a field talking to itself in a venue that counts.

Benchmarks flatter; portfolios decide

The review's sharpest policy recommendation: stop treating model validation as the finish line. Measure whether AI changes decisions — which targets survive triage, which molecules get synthesized, which trials get funded.

August's agentic headlines — Schrödinger Bunsen at BMS, Phylo Biomni at Chugai — are exactly the kind of workflow claims that must be judged on portfolio outcomes, not press releases.

Stephen Johnson at BMS said AI lets scientists scale creativity. Fine — show the Phase II choices Bunsen altered, not just the workflows it ran.

Technology push is not science pull

The Perspective names an underlying dynamic: technology push vs. science pull. Vendors ship capability; discovery teams need conditional, messy, decision-grade tools.

Insilico's rentosertib Phase III (initiated July 7) is the counterexample everyone cites — and the authors would rightly note one program does not validate a field.

Why August 7 timing stings

The same week Anthropic widened Fable 5 biology access and OpenAI flagged Critical cyber risk for Astra, a Nature Reviews paper reminds us that biological capability without clinical proof is a governance problem too — not only a misuse problem.

Wider model access for benign biology helps patients-in-waiting only if discovery pipelines convert capability into validated medicines.

What would change my mind

• Published case studies linking AI recommendations to prospective experimental outcomes
• Benchmarks reporting decision delta, not AUROC alone
• Pharma ML teams sharing negative results when models did not change choices

Closing view

The AI drug-discovery field does not need less ambition. It needs honest accounting — and Nature Reviews just provided the template. Celebrate methods; demand decisions.

### Sources

• Nature Reviews Drug Discovery — Artificial intelligence in drug discovery — what it is, where we stand and the path forward (August 7, 2026)
• Schrödinger — Strategic Collaboration with Bristol Myers Squibb to Deploy Bunsen (August 6, 2026)
• EurekAlert — Insilico initiates Phase III clinical trial for rentosertib (July 7, 2026)
• Anthropic — Improving Fable 5's biology safeguards (August 7, 2026)