CEO Confronted Over Using AI to Clone Real People Without Their Consent

Superhuman (formerly Grammarly) discontinued its 'Expert Review' AI feature after public backlash and has been hit with a class-action lawsuit led by journalist Julia Angwin alleging the company trained AI on writers' work and used their names without permission. CEO Shishir Mehrotra apologized but defended the product as attribution rather than impersonation, saying usage was low; the company maintains the legal claims are without merit. Reputational and litigation risk is moderate and could spur regulatory or legal scrutiny of AI content tools, but the story is unlikely to move markets materially in the near term.

Analysis

The episode accelerates a bifurcation: consumer-facing AI assistants that monetize broad, unlicensed training sets will face rising legal and reputational costs, while enterprise-hosted, consent-first model deployments gain pricing power. Expect indemnity, licensing and provenance controls to become budget line-items — conservatively add 1–3% of SaaS revenue to legal/compliance spend for exposed consumer AI vendors within 12 months, and a 2–5% lift in cloud/compute spend for customers migrating to hosted, auditable models over 6–18 months.

Second-order winners are products that enable attribution, content provenance, and identity protection: watermarking, signed-model outputs, content-licensing marketplaces, and identity-verification services will see both demand and new revenue streams (subscription and per-transaction fees). Conversely, small independent model providers and consumer apps that cannot cost-effectively underwrite IP risk will either consolidate or exit — accelerating consolidation of model hosting with large cloud vendors and platform providers within 12–24 months.

Key catalysts to watch are legal rulings and regulatory guidance (FTC, EU AI Act, U.S. copyright clarifications) over the next 6–24 months; a plaintiff win or a precedent-setting settlement would force accelerated licensing markets and insurer repricing. The thesis can be reversed quickly if courts broadly reject claims or if credible industry-wide licensing/consent marketplaces launch and scale in 3–6 months, which would re-open the low-cost consumer assistant market and compress margins for provenance vendors.

Key Decisions for Investors

• Long CRWD (CrowdStrike), 6–12 months: buy shares or 6–12 month call spreads. Rationale: identity/endpoint demand rises as firms harden model access and attribution; asymmetric upside if enterprise security budgets reallocate 2–4% to model governance. Risk: macro headwinds and multiple compression; target 20–35% upside vs 15–20% downside.
• Long MSFT or GOOGL, 6–18 months: buy shares or buy-write collars. Rationale: migration to enterprise-hosted, auditable models increases Azure/GCP IaaS & PaaS consumption by an estimated incremental 2–5% of cloud spend for large customers. Risk: competition and margin pressure from open-source; expected reward 15–25% vs 10–15% downside.
• Long RELX (RELX plc) or TRI (Thomson Reuters), 9–18 months: selective longs in legal/IP data providers. Rationale: surge in licensing/dispute management increases demand for IP enforcement analytics and content-licensing tooling; payoff durable post-legal rulings. Risk: slower adoption if litigation settles quietly; target 25%+ upside vs 12% downside.
• Tactical hedged short of consumer AI app group / pair trade, 3–9 months: short a basket of consumer-facing AI writing assistants (small-cap or weak-balance-sheet names) while longing enterprise SaaS peers (CRWD/MSFT). Rationale: regulatory/legal shock will compress valuations of under-provisioned consumer apps faster than enterprise peers. Risk: headlines or quick settlements that restore consumer confidence; maintain size caps and 2:1 hedge ratio.