AI Token Securities Analysis: The Howey Test in 2026

Every AI token project eventually makes the same argument to its lawyers: our token isn’t a security, it’s how users pay for inference. It is a real argument. It is also, standing alone, a losing one. Under SEC v. W.J. Howey Co., 328 U.S. 293 (1946), consumptive utility is a fact in the analysis — not a defense to it. For teams building on the AI and blockchain frontier, the operative question in 2026 is not whether the token does something useful, but whether the federal securities laws treat the purchase of that token as an investment in someone else’s ongoing work.

That question is unusually hard for AI tokens, and the reason is structural. AI networks depend on concentrated, continuous, expensive human effort — model training, data curation, inference infrastructure, safety tuning — in a way that a payments chain or an automated market maker does not. The fourth Howey prong asks precisely about that dependency. AI architecture, in other words, pushes toward securities treatment at the exact place most token issuers assume they are safe.

Why AI Token Securities Analysis Starts With Distribution, Not Technology

The Securities Act defines “security” to include any “investment contract” (15 U.S.C. § 77b(a)(1)), and Howey supplies the test: an investment of money in a common enterprise with a reasonable expectation of profits derived from the efforts of others (328 U.S. at 298-99). Critically, Howey examines the transaction, not the asset. The same AI token can be sold as a security on Monday and traded as something else on Tuesday.

That distinction drove the most consequential district court ruling in crypto. In SEC v. Ripple Labs, Inc., 682 F. Supp. 3d 308 (S.D.N.Y. 2023), the court held that institutional sales satisfied Howey while blind bid/ask exchange sales did not, because anonymous secondary buyers could not know their money reached the issuer. That holding has never received appellate endorsement, and practitioners should not treat it as settled law — but its logic is directly relevant to AI projects, most of which distribute through a mix of private rounds, emissions, and open-market trading.

The Division of Corporation Finance’s Framework for “Investment Contract” Analysis of Digital Assets (April 3, 2019) remains the most detailed public articulation of staff thinking. It is non-binding, it predates the current Commission, and it has not been updated for AI — but its factors read like a checklist written for AI token cap tables: an “Active Participant” essential to the network, a network still under development, tokens sold broadly rather than to expected users, and promotional emphasis on appreciation.

Trained by Whom? The Fourth Prong Problem

Here is the analysis most AI projects have not run. A decentralized exchange can credibly claim that, post-launch, immutable contracts run without anyone. An AI network usually cannot. Someone retrains the model. Someone curates the dataset. Someone pays the GPU bill and decides which version serves production traffic. Under the 2019 Framework, that someone is an Active Participant, and purchasers who expect the token to appreciate as the model improves are expecting profits from that participant’s efforts.

The counterargument — that permissionless compute markets and subnet-style incentive mechanisms distribute those efforts across independent operators — is legitimate and, in some architectures, strong. But it is an argument about facts, and it has to be true at the level of the specific network. The following factors, drawn from the Framework and applied to AI-specific architecture, are how we would expect a court to sort it out. These are analytical distinctions, not agency doctrine; the SEC has not litigated an AI-utility token to judgment.

The buyback-and-burn row deserves emphasis. A burn funded by protocol revenue is economically a distribution to holders. Pairing it with marketing about scarcity is close to the strongest profit-expectation evidence a project can manufacture against itself.

Architectures across the sector vary widely on these axes — Bittensor-style subnet emissions, Render and Akash compute marketplaces, and agent frameworks like Fetch.ai or Ocean Protocol each present materially different facts. We reference them as illustrations of design variation only, and express no view on the classification of any specific token.

Utility Does Not Cure the Offering

SEC v. LBRY is the case AI issuers should read first. LBRY’s token had a functioning consumptive use inside a live content network. The court granted summary judgment for the SEC anyway, reasoning that the objective economic realities of LBRY’s offering — including messaging about the token’s value growing with the network — would lead a reasonable investor to expect profits from LBRY Inc.’s efforts. Utility coexisted with the investment contract; it did not displace it.

SEC v. Telegram Group Inc., 448 F. Supp. 3d 352 (S.D.N.Y. 2020), adds the second lesson: courts assess the entire scheme, including the anticipated resale into public markets, rather than each contractual step in isolation. An AI project that sells to accredited purchasers under Rule 506(c) while planning a broad public listing is running Telegram’s playbook.

None of this means AI tokens are categorically securities. It means the analysis is transaction-specific and fact-intensive, and that the Howey test factors reward projects that ship working product before raising against a roadmap.

Where AI Regulation and Crypto Regulation Collide

US issuers with EU users now face a compliance interaction that did not exist two years ago. Regulation (EU) 2024/1689 — the EU AI Act — phased in on the schedule set by Article 113: prohibited practices under Article 5 from February 2, 2025; general-purpose AI model obligations under Articles 53 and 55 from August 2, 2025; and high-risk classification under Article 6 and Annex III, plus the Article 50 transparency duties, from August 2, 2026. Article 25 extends responsibility along the value chain, so a deployer that substantially modifies a model can itself become a provider.

The securities consequence is counterintuitive. Article 3(3) requires an identifiable provider — a named legal person responsible for the model. A project that appoints an EU-facing provider entity, maintains technical documentation, and controls model releases has created a documented record of centralized managerial control. That record is discoverable, and it is exactly what a plaintiff or the SEC would use to rebut a decentralization defense under Howey’s fourth prong. This cross-regulatory tension is our analysis; no agency has addressed it.

Separately, Regulation (EU) 2023/1114 (MiCA) imposes white paper and notification duties under Articles 4 and 5 for public offers of crypto-assets other than asset-referenced or e-money tokens, subject to exemptions in Article 4 covering, among others, tokens automatically created as a reward for validating transactions and utility tokens giving access to a good or service that already exists. US firms that block EU access typically rely on Regulation S; the offshore-transaction and directed-selling-efforts conditions are easy to breach with a public Discord.

The Second Liability Vector: AI Washing

Classification is not the only exposure. On March 18, 2024, the SEC settled parallel actions against Delphia (USA) Inc. and Global Predictions Inc. for misrepresenting their use of artificial intelligence, charging violations of the Investment Advisers Act (15 U.S.C. § 80b-6) and the Marketing Rule, 17 C.F.R. § 275.206(4)-1. The principle generalizes: Section 17(a) of the Securities Act and Rule 10b-5 reach material misstatements in connection with the offer or sale of a security, and if the token is not a security, the CFTC’s manipulation and fraud authority under 7 U.S.C. § 9(1) and Regulation 180.1 may apply instead. Describing a fine-tuned wrapper around a third-party foundation model as “proprietary AI” is a disclosure problem under every one of those regimes.

What to Do Now

1. Paper the distribution, not just the token. Build a transaction-by-transaction record — private rounds, airdrops, emissions, liquidity provisioning — and run the Howey analysis separately for each. Ripple‘s distinction, whatever its precedential fate, tracks how courts think about issuer proceeds.
2. Audit your public statements for profit language. Pull every deck, thread, blog post, and AMA. Scarcity claims, burn mechanics, staking-yield marketing, and “value accrual” framing are the evidence LBRY turned on.
3. Document decentralization as of a date, with specifics. Count independent inference operators, measure the share of traffic not touching founder infrastructure, and record whether weights are forkable. A conclusory “we’re decentralized” memo is worth nothing.
4. Reconcile your EU AI Act filings with your US securities position. Have the same lawyer read both. If the provider documentation says one entity controls the model while the securities memo says no one does, you have a problem before anyone subpoenas you.
5. Substantiate every AI capability claim. Maintain evidence for what the model is, what it was trained on, and what it actually does. Antifraud liability attaches regardless of classification.
6. Assume no retroactive fix. Any future exemptive framework is more likely to govern prospective conduct than to cleanse completed unregistered offerings. Our crypto compliance guide covers the remediation options that exist today.

What to Watch

Three open threads. First, federal market structure legislation: the Digital Asset Market Clarity Act (H.R. 3633) passed the House in July 2025, and its “mature blockchain system” concept would, if enacted in that form, give decentralization a statutory definition rather than a factual one. Readers should verify its current status, which may have changed since drafting. Second, the SEC’s Crypto Task Force, announced January 21, 2025 under Commissioner Hester Peirce, has explored a token taxonomy and a time-limited exemptive path but had not produced binding rule text as of our last confirmed review — and staff statements are not Commission rules. Third, on the AI side, the Commission’s 2023 proposal on conflicts of interest in predictive data analytics (Rel. No. 34-97990) was proposed and, we understand, subsequently withdrawn; any successor rulemaking would land squarely on AI-driven DeFi and trading systems.

Until those threads resolve, AI token securities analysis runs on 1946 law applied to 2026 architecture. That is uncomfortable, but it is not ambiguous about what to do: assume Howey applies, build the factual record that supports your position, and stop marketing utility tokens with appreciation language.

Sources & References

1. SEC v. W.J. Howey Co., 328 U.S. 293 (1946)
2. Securities Act of 1933, 15 U.S.C. § 77b(a)(1)
3. SEC Division of Corporation Finance, Framework for “Investment Contract” Analysis of Digital Assets (April 3, 2019)
4. SEC v. LBRY, Inc., 639 F. Supp. 3d 211 (D.N.H. 2022)
5. SEC v. Telegram Group Inc., 448 F. Supp. 3d 352 (S.D.N.Y. 2020)
6. SEC v. Ripple Labs, Inc., 682 F. Supp. 3d 308 (S.D.N.Y. 2023)
7. Regulation D, Rule 506(c), 17 C.F.R. § 230.506(c)
8. Regulation S, Rule 903, 17 C.F.R. § 230.903
9. SEC Press Release 2024-36, Charges Against Delphia (USA) Inc. and Global Predictions Inc. for AI-Related Misstatements (March 18, 2024)
10. Investment Advisers Act § 206, 15 U.S.C. § 80b-6
11. Rule 10b-5, 17 C.F.R. § 240.10b-5
12. CFTC Regulation 180.1, 17 C.F.R. § 180.1
13. Regulation (EU) 2024/1689 (EU Artificial Intelligence Act)
14. Regulation (EU) 2023/1114 (Markets in Crypto-Assets Regulation)
15. Digital Asset Market Clarity Act of 2025, H.R. 3633 (119th Congress)

This article is for informational and educational purposes only and does not constitute legal advice. Reading this content does not create an attorney-client relationship. Laws and regulations in the cryptocurrency space change rapidly — consult a qualified attorney for advice specific to your situation and jurisdiction. Seth Goettelman is licensed to practice law in Wisconsin and Florida.

Building or advising an AI token project? Our consultation page explains how a transaction-level securities review works. For ongoing coverage of the AI and crypto regulatory intersection, subscribe below.