How AI Could Freeze Progress with Hilary Allen

What It Covers

Law professor Hilary Allen, author of Fintech Dystopia, examines how Silicon Valley's venture capital model uses regulatory arbitrage rather than genuine technological innovation to build financial products. Allen connects lessons from the 2008 financial crisis to current crypto, AI, and fintech trends, arguing deregulation cycles repeat because institutional memory fades predictably fast.

Key Questions Answered

• •Regulatory Arbitrage as Core Business Model: Most fintech value comes from legal maneuvering, not technical superiority. Crypto exchanges like Coinbase combine broker-dealer and exchange functions — explicitly prohibited under securities law — and survive by lobbying for carve-out legislation rather than building genuinely superior products. Incumbents must comply with all regulations while disruptors operate outside them, creating an unlevel playing field that rewards legal strategy over product quality.
• •LLM Accuracy Has Hit a Ceiling: Large language models are statistical engines that predict probable word sequences without any capacity to verify accuracy as a concept. Allen argues these models have reached their accuracy ceiling and will not improve further on hallucinations. The practical risk: professionals reviewing AI-generated work are statistically more likely to miss errors than if they had written the content themselves, creating compounding malpractice exposure.
• •Venture Capital Subsidies Distort Market Selection: Standard market logic — where inferior products fail — does not apply to VC-backed fintech because of structural subsidies: ERISA rule changes allowing pension fund investment in VC, carried interest tax treatment, and access to cheap capital in low-rate environments. Crypto should have collapsed in the 2022 winter but Andreessen Horowitz redirected its war chest toward congressional lobbying instead of new investments.
• •Innovation Rhetoric Follows a Predictable Playbook: Framing products as novel and unregulatable is a deliberate narrative strategy, not an accurate technical description. Fintech lending replicates predatory payday lending structures with equivalent effective interest rates — sometimes exceeding 300% annually when fees are annualized. Buy-now-pay-later products avoid lending disclosure laws by rebranding loans as fee-based arrangements, exploiting the same economic precarity that made payday lending profitable.
• •AI Freezes Knowledge Accumulation Over Time: As AI tools replace junior lawyers, writers, and analysts, the humans who would generate new case law, briefs, and creative work disappear from the pipeline. LLMs have a fixed training cutoff and cannot incorporate new developments autonomously. Younger professionals who never develop independent skills cannot spot AI errors, creating a generation unable to audit the tools they depend on — a structural degradation of institutional knowledge.
• •Deregulation Cycles Are Driven by Memory Loss, Not Evidence: Regulatory frameworks erode predictably once the crisis that created them recedes. Securities investor protections built after the 1930s crash are currently being dismantled. The 2008 crisis produced Dodd-Frank only in 2010, by which point industry lobbying had weakened structural reforms. Allen identifies the same pattern now: post-crisis regulatory consensus dissolves within roughly two years as political economy shifts back toward incumbent financial interests.