Letters: Pharma bets, judicial AI and the limits of the machine

Three items this week converge on a single structural question: what happens when pharmaceutical capital, AI litigation tools, and state involvement in drug development start moving in the same direction at the same time?

Eli Lilly stacks its pipeline

Eli Lilly & Co. announced on 26 May 2026 that it would acquire three clinical-stage vaccine developers in a deal valued at up to $3.8 billion. The transaction — the kind of bolt-on acquisition that large-cap pharmaceutical companies use to shorten development timelines — expands Lilly's immunisation portfolio at a moment when the broader vaccine market has attracted renewed investor interest following a series of mRNA-era partnerships across the industry.

The timing matters. Also on 26 May, prediction markets were assigning a 27 percent probability to the United States taking a stake in Eli Lilly — a scenario that would represent a significant departure from the conventional relationship between the federal government and the pharmaceutical sector. Normally, federal investment in drug development flows through mechanisms like BARDA (the Biomedical Advanced Research and Development Authority) or through advance purchase agreements, not through equity ownership of a listed company. A 27 percent probability is not a forecast; it reflects traders assigning meaningful weight to a non-standard outcome. Whether that reflects genuine policy intelligence or speculative positioning on regulatory arbitrage is not clear from the market data alone.

Separately, Eli Lilly confirmed it is developing a one-time gene-editing injection that the company claims could permanently lower LDL cholesterol — so-called bad cholesterol — in a single administration. If the clinical data holds, the product would represent a competitive threat to established PCSK9 inhibitors (sold by Regeneron and Amgen) and to the siRNA-based approach taken by Novartis with inclisiran. The claim to permanence is the clinically significant variable: current lipid-lowering therapies require ongoing dosing, and patient adherence fatigue is a documented problem across the cardiovascular drug class. A one-shot solution, if safe and durable, could upend the economics of a multi-billion-dollar maintenance market.

The British court and the hallucinating lawyer

In what legal commentators are already describing as a landmark ruling, a British judge found on 26 May 2026 that a lawyer had "almost entirely outsourced the thinking process" to an AI tool after submitting court documents containing hallucinated citations — references to case law that did not exist. The ruling marks one of the first judicial rebukes specifically addressing the delegation of legal reasoning to large language models without adequate human review.

The practical stakes are straightforward. Court systems in the United Kingdom, the United States, and across the European Union are already processing a rise in AI-drafted filings. Most jurisdictions have responded with disclosure rules — requiring lawyers to confirm that AI tools were used in document preparation — but enforcement mechanisms remain inconsistent. The British ruling does not create binding precedent for other jurisdictions, but it signals that judicial gatekeepers are willing to distinguish between AI as a drafting aid and AI as a substitute for legal judgment. The distinction matters: one is a productivity tool; the other is a liability.

Hallucination in legal citation is a known failure mode of large language models and one that existing hallucination-detection tools have addressed imperfectly. What the ruling surfaces is the gap between tool capability and professional responsibility. A document that looks polished and is grammatically coherent can still contain fabricated authority — and a court that relies on it can reach an incorrect judgment. The lawyer in this case did not appear to have verified the citations. That failure, not the use of AI per se, is what the judge found problematic.

The structural pattern

Pharmaceutical consolidation, gene-editing pipelines, and AI-adjacent legal disputes do not share a clinical cause. But they share a governance problem: each operates in a zone where the rules governing responsibility are lagging well behind the capabilities being deployed. The US government's potential equity stake in Eli Lilly — speculative as it remains — sits alongside existing precedents for state involvement in strategic industries. The British judge's ruling sits alongside a body of professional guidance on AI use in law that has yet to crystallise into binding conduct rules.

What is consistent across both is a rhythm of deployment outpacing accountability. Drug companies are acquiring clinical-stage assets and announcing gene-editing breakthroughs before regulatory frameworks for pricing, safety monitoring, and indication expansion have fully adapted to the speed of the pipeline. Lawyers are using AI tools whose output they cannot fully verify, in systems where verification is the entire point of the professional obligation. And markets are pricing the uncertainty — 27 percent on a US equity stake in one of America's largest pharmaceutical companies is not a small wager on regulatory novelty.

The broader question is not whether these technologies or transactions are good or bad. It is whether the institutional infrastructure — courts, regulators, disclosure norms — is moving fast enough to allocate accountability when something goes wrong. On current evidence, the answer is no. And the British judge's terse finding — that a lawyer delegated their thinking to a machine — may end up being the most direct description of the underlying problem yet offered by a court.

This publication covered Eli Lilly's vaccine acquisition and the AI legal ruling from Polymarket-sourced wire reports. The gene-editing claim and the Polymarket probability on a US equity stake in Eli Lilly remain subject to confirmation from primary corporate disclosures.