Jason Riley argues the dominant AI risk narrative — job losses, deepfakes, existential threat — is a policy trap. The greater danger, he writes, is that fear-driven licensing regimes will kneecap the technology that is already, quietly, lifting wages in the skill tiers where it is allowed to run.
Riley cites career coach Gina Riley's (no relation) observation that deepfake risk is real but localized — and that productivity gains in mid-skill white-collar work are broader and more durable than the loss-of-jobs story admits.
The Licensing Trap
Riley's argument mirrors what William Galston wrote on A15 about the upper-middle-class pulling away: if AI is only deployable by firms large enough to fund a compliance department, the productivity dividend stays inside the top decile. Small business owners, career-switchers, and solo professionals lose their one real accelerant.
The Foolish, Fear-Driven Policies
Riley names three: (1) sector-by-sector licensing regimes with no reciprocity across states, (2) training-data liability rules that penalize open-source, and (3) export controls framed as safety but functioning as industrial protection.
What This Means For Our Portfolios
- AI/Cyber 20% — tilt toward tooling. MSFT, GOOGL, NVDA. But also PLTR, SNOW, DDOG which sit in the productivity-layer Riley is defending.
- Watch open-source exposure. META's Llama position, IBM Granite. If liability rules pass, closed-weight incumbents (MSFT, GOOGL) gain; open-source loses.
- Political tail: state-by-state licensing. A CA or NY licensing bill would hit mid-cap SaaS harder than mega-caps — watch HUBS, TEAM, ZS.
- For client conversations: Riley is the antidote to the Véliz review on the same page's sister column. Véliz warns about algorithmic determinism; Riley warns about regulatory over-correction. Both risks are real — diversify across them.