What happened
AI-designed drug candidates are clearing Phase I clinical trials at 80 to 90 percent, well above the historical industry rate of 40 to 65 percent. However, of 117 tracked AI-enabled assets in clinical trials, only 8 have cleared Phase II—where efficacy in actual patients is tested. The Phase II success rate remains at roughly 40 percent, unchanged from decades past.
Why it matters
The industry has celebrated Phase I clearance as proof that AI works in drug discovery, but Phase I only asks whether a molecule is safe to test in humans—not whether it actually treats disease. According to Bristol Myers Squibb's Chief Digital and Technology Officer, AI is "moving rather than removing bottlenecks": it generates candidate molecules more efficiently, but the real constraint—validating which biological mechanism actually changes disease in living patients—still requires expensive, slow human trials. Most AI investment has chased the faster, measurable wins (Phase I) rather than the harder, slower validation (Phase II).
What to watch
Whether the first wave of Phase III readouts in 2026 will change the picture. Per a ZS 2026 survey of 115 pharma and biotech digital leaders, 68 percent cite poor data quality and governance, not AI model capability, as the reason AI initiatives stall—suggesting the bottleneck is trustworthy training data, not the algorithms themselves.
Summaries like this, in your inbox every morning.
The article lays out a paradox at the heart of AI-in-pharma hype: the industry has celebrated a genuine win (Phase I clearance rates rising from 40–65% to 80–90%) as proof of concept for AI-driven discovery, but that victory reveals a category mistake. Phase I is a tolerability screen—a necessary but narrow gate. The question that matters for patient benefit—whether a molecule actually treats disease—is answered in Phase II, where success rates have not budged in decades.
The structural reason, according to voices like Greg Meyers (Bristol Myers Squibb) and Vikram Singh (Gilead Sciences), is that AI has excelled at the wrong optimization target. AI generates candidate molecules efficiently from well-understood biology, but it has not removed the real bottleneck: validating which biological mechanisms actually alter disease in living humans. That validation still requires randomized clinical trials—the slowest and most expensive experiment in science. In other words, AI moved the bottleneck upstream (making candidate generation faster) but did not remove the downstream bottleneck (proving efficacy in patients). The industry's investment narrative followed the faster signal: Phase I clearance is measurable and reportable. Phase II validation is slow, expensive, and humbling.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
OpenAI introduced the Rosalind Academic Plan, giving eligible academic labs at U.S
Wells Fargo hired an AI executive based in San Francisco, a move The Business Journals describes as part of em…
Amazon launched two Bee "personal AI" devices on Thursday: the $74.99 Bee Frontier wristband and the $99.99 Be…
Apple released MLX today, which runs the 816B-parameter Muse Spark Open on four Mac Studios at over 36 tokens…
Bloom Energy stock fell 4% to $229.83, still up 165% year to date, while Caterpillar slipped 3% to $727.55 and…
LogicMonitor announced its 2026 Elevate Award winners at its Elevate Community Conference in Chicago, honoring…
