- Nate raised $42M+ for an "AI-powered" shopping app. Real automation rate? Essentially zero — hundreds of contractors in the Philippines manually completing every purchase. April 2025: the SEC and DOJ charged the founder. Up to 20 years in prison. 🧵
- This is "AI washing": claiming AI does work that humans (or nothing) actually do. It stopped being a marketing problem. The SEC, FTC, and state AGs now prosecute it with the same tools they use for securities fraud.
- The first cases landed March 2024: Delphia ($225K) and Global Predictions ($175K), charged under the Advisers Act and Marketing Rule for AI claims they couldn't substantiate. $400K combined. The floor was set.
- Then it escalated. Jan 2025: Presto Automation — the first AI-washing case against a public company. It claimed AI eliminated human order-taking. The SEC found 70%+ of orders still needed humans. 100% at some locations.
- Feb 2025 the SEC stood up a dedicated unit: CETU, the Cybersecurity and Emerging Technologies Unit, built to pursue AI misconduct. Its 2026 exam priorities now actively sweep RIAs and broker-dealers for AI washing. This is structural, not a one-off.
- Every action runs the same play: compare what you SAID about your AI to what it actually DOES. Delphia never integrated the data it touted. Presto needed humans. Nate's automation was zero. The gap is the case.
- It's not just the SEC. The FTC's Operation AI Comply nailed Workado for advertising a "98% accurate" AI detector that tested at ~53%. State AGs now wield UDAP statutes and new laws — Colorado at $20K/violation, NY at $15K/day.
- The plaintiffs' bar smells blood too: 53 AI-related securities class actions filed through H1 2025 (Stanford Law). Median settlement $11.5M. Tucker v. Apple even targets the Apple Intelligence promises made at WWDC.
- Most AI governance platforms sell dashboards. An SEC examiner doesn't want a dashboard — they want the evidence chain: proof your AI does what your 10-K says. Stanford RegLab clocked LLMs hallucinating 69-88% on legal queries; "the model said so" isn't a defense.
- That's our work: AI verification architecture, AIBOMs (AI Bills of Materials), and claim-substantiation packages that hold up when a regulator says "prove it." Not governance theater — the actual defensible evidence.
- If a regulator asked tomorrow for proof your AI does what your marketing claims, could you produce the documentation — or just a dashboard? #AIGovernance #RegTech
- We mapped the full enforcement landscape — SEC, FTC, state AGs, EU AI Act — and what defensible substantiation actually takes: https://veriprajna.com/solutions/ai-verification-anti-ai-washing