In September 2026, Anthropic CEO Dario Amodei posted on social media that it is "perhaps possible" for AI to cure most human diseases within the next 5 to 10 years. Around the same time, another tech executive made an even bolder prediction: AI will cure 90% of human diseases within 7 years. These statements sparked widespread discussion in tech circles, but in the pharmaceutical industry, executives in charge of clinical pipelines gave a completely different answer.
According to a Wall Street Journal report on November 4, 2026, Vertex Pharmaceuticals Executive Chairman Jeff Leiden described an internal test: the company invited about 50 AI companies to analyze a dataset containing 10,000 drug compounds. The result: not a single AI company produced noteworthy results. Faced with Silicon Valley's assertions that AI is about to conquer disease, Leiden's assessment was direct and brief: "That's nonsense."
As of the report date, no AI-designed drug anywhere in the world had received marketing approval from the U.S. Food and Drug Administration (FDA).
Silicon Valley Is Betting with Action, Pharma Is Responding with Data
Tech companies' optimism is not just talk. Anthropic has quietly acquired two early-stage biotechnology companies; OpenAI acquired a startup focused on AI drug discovery in October 2026.
AstraZeneca CEO Pascal Soriot said: "Disease is complex, biology is complex. Saying AI will cure cancer may be a bit of a stretch." Regeneron President and Chief Scientific Officer George Yancopoulos was more specific: AI will assist at the laboratory level, but will not "miraculously tell us what to do."
What AI currently excels at is high-dimensional pattern matching—target identification, molecular screening, compound optimization—tasks whose inputs and outputs can be structured. But for a drug to complete the entire journey from hit compound to market, it must traverse countless unknown "dark matter" territories of human biology, and existing AI models have almost no weapons in this area.
How Complex Drug Development Failure Is: The 25% AI Doesn't Know
McKinsey Senior Partner Alex Deverson provided a specific quantitative perspective: among drugs entering mid-to-late-stage clinical trials, about 25% of failures are due to "off-target" safety issues—the drug acts on other parts of the body while targeting the intended goal, causing side effects. According to his analysis cited by the Wall Street Journal, current AI models have limited ability to predict such "off-target" effects.
The cost of drug R&D failure is measured in billions of dollars, and failures tend to concentrate in the most expensive clinical stages. If AI cannot provide early warnings of "off-target" risks, the time saved in early screening may be offset by later clinical losses. This also explains why pharma executives generally view AI as a tool to accelerate early research rather than a lever to shorten the entire R&D cycle.
Big Money Is Flowing, but the Bet Structure Is Different
Pharma's caution is not reflected in capital retreat. AstraZeneca reached a collaboration with AI biotech company Insitro, paying $40 million upfront with milestone payments of up to $1.46 billion, focused on new therapies for liver and heart disease. The deal structure reveals the real logic of the pharmaceutical industry: using small upfront investments to buy options on future possibilities, rather than treating "curing disease" as a certain path the way tech companies do.
Tech companies directly acquire biotech firms, while pharma companies retain exit room through milestone-based contracts. Behind these two capital behaviors are different judgments about AI's role in drug R&D: the former bets AI will disrupt the R&D paradigm, while the latter sees AI as an accelerator within existing processes, not a replacement.
Validation Window: FDA's First AI-Designed Drugs Are the Key Signal
The most direct validation checkpoint is the FDA approval record. If within the next two to three years, a molecular compound designed primarily by AI (not merely optimized with assistance) passes Phase III clinical trials and receives marketing approval, that would provide substantive support for Silicon Valley optimists' arguments; if this gap persists, pharma executives' technological conservatism will gain more support.
The second signal is the renewal status of collaborations between pharma companies like Vertex and AI firms. The case of Vertex testing 50 AI companies shows that its openness to AI technology is genuine, but its evaluation standard is actual results, not demos and papers. If the next batch of collaborations produces repeatable positive outputs, executives' public statements will quickly follow.
The third observable dimension is the specific progress of OpenAI and Anthropic in biotechnology. After completing their acquisitions, whether the two companies can produce drug candidate compounds reaching the preclinical stage within the next 12 to 24 months will be an early test of whether their strategic intentions in medicine are executable.
Pharma executives' pushback is not a denial of AI's value to the industry, but a correction of the timescale and narrative boundaries. In specific areas such as target screening, patient stratification, and clinical trial design, AI is already changing how work is done. But from "accelerating a certain step" to "curing disease," what lies in between is the entire unknown territory of human biology.
© 2026 Winzheng.com 赢政天下 | 转载请注明来源并附原文链接