Who Holds the Keys.

Rev. Proc. 2026-20 is written for a narrow audience: exchange-listed products that hold a single proof-of-stake asset. Almost no family trust will ever qualify for it, and none needs to. It earns a place in this series for a different reason. In working out when a trust may stake without ceasing to be a trust, the Treasury Department and the IRS had to write down what they think separates a trust from a business. Their answer turns on who controls the asset, how much discretion the trustee has, and how far the trust stands from the people it pays. Those are the same questions a creditor’s lawyer asks about a private trust, for different reasons and under different law.

Federal tax law has classified trusts by function for a long time. An arrangement is treated as a trust, the procedure reminds us, if its purpose is to vest in trustees the responsibility “to protect or conserve property for beneficiaries who cannot share in the discharge of this responsibility” (Id. § 3.02, describing Treas. Reg. § 301.7701-4(a)). An arrangement that uses the trust form to carry on a profit-making business is classified as a business entity instead (Id. §§ 3.01, 3.03, describing Treas. Reg. §§ 301.7701-2(a) and 301.7701-4(b)).

A pooled investment trust sits close to that line, so the regulations give it a sharper test. An investment trust keeps its trust classification only if there is no power under the trust agreement to vary the investment of the certificate holders (Id. § 3.04, describing Treas. Reg. § 301.7701-4(c)). The procedure, citing the Second Circuit’s 1941 decision in Commissioner v. North American Bond Trust, describes that forbidden power as “a managerial power under the trust instrument that enables a trust to take advantage of variations in the market to improve the investments of the certificate holders” (Id. § 3.05, citing Commissioner v. North American Bond Trust, 122 F.2d 545 (2d Cir. 1941), cert. denied, 314 U.S. 701 (1942)).

Staking strains both tests. A trust that commits its tokens to a validator, earns rewards, chooses providers, and decides how much to stake and when is doing more than holding. It can look like a trustee managing for yield, or like a business. In 2025 sponsors asked the Treasury and the IRS whether staking would cost a trust its classification, and on November 10, 2025 the agencies answered with Rev. Proc. 2025-31 (Id. § 2.08; Rev. Proc. 2025-31, 2025-48 I.R.B. 743). Practitioners then came back with questions on eight points, among them which protocols were covered, whether the SEC had to “approve” disclosures, whether more than one custodian could be used, and how much slashing protection was required (Rev. Proc. 2026-20, § 2.09 (listing eight subjects on which additional clarity was requested)). Rev. Proc. 2026-20 is the reply. It clarifies, modifies, and supersedes the 2025 procedure (Id. §§ 1, 9).

The procedure applies to an arrangement formed as a trust under State law that already qualifies as an investment trust and as a grantor trust (Id. § 5 (status “as determined immediately before satisfying all the requirements in section 6.02”)). If all fourteen requirements of section 6.02 are met, the trustee’s authority to stake and the staking itself “do not prevent the trust from qualifying” under either heading (Id. § 6.01). The requirements fall into five groups.

A listed, single-asset vehicle. Interests in the trust trade on a national securities exchange, the staking disclosure sits in an effective SEC registration statement, and the trust owns only cash and units of one digital asset on a permissionless proof-of-stake network (Id. § 6.02(1)–(2)).

Custody. The assets are held by one or more custodians at addresses the custodians control. Only the custodian can reach the private keys, so only the custodian can sell, transfer, or exercise ownership rights, including while the assets are staked. For tax purposes the trust “retains ownership of the digital assets at all times, including while they are staked” (Id. § 6.02(3)).

A trustee with a short list of powers. The trustee may accept deposits, hold, pay expenses, distribute, liquidate, and direct staking. The trust agreement must prohibit the trustee “from seeking to take advantage of variations in the market to improve the investments of trust interest holders, including variations based on the value of the digital assets or the amount of staking rewards” (Id. § 6.02(5)). Subject to defined exceptions for liquidity reserves, short-term operational holds, and protective measures, all of the trust’s assets must be made available for staking at all times, which removes the timing decision from the trustee (Id. § 6.02(8)–(12)). Staking itself is characterised as conservation: it “protects and conserves trust property by mitigating the risk that another party or group could control a majority of the total staked digital assets of that type” (Id. § 6.02(4)).

Distance from the staking provider. The trust and its sponsor must be unrelated to the staking provider. The provider must regularly do similar work for unrelated persons, bear its own expenses, and split rewards on an arm’s-length allocation (Id. § 6.02(6)). The trust, the custodian, and the sponsor may have “no legal right or arrangement to participate in or direct or control the activities of the staking provider in any way,” other than directing staking and unstaking (Id. § 6.02(7)).

Loss protection and no accumulation. The trust must be indemnified “against slashing due to activities or events reasonably within the staking provider’s control or ability to protect against” (Id. § 6.02(13)). The only new property staking may bring in is more units of the same asset, and net rewards must go out to holders, in kind or in cash, “no more than 60 days after the end of the calendar quarter in which the trust gains dominion and control over the relevant staking rewards” (Id. § 6.02(14)).

Three of the revisions are worth knowing. The 2025 text required that the trust’s staking disclosure had “been reviewed and approved by the SEC”; the new text asks that it be filed in an effective registration statement and remain subject to the SEC’s oversight (compare Rev. Proc. 2025-31, § 6.02(1), with Rev. Proc. 2026-20, § 6.02(1)). The 2025 text spoke of “a custodian”; the new text says “one or more custodians” and ties each custodian’s exclusive key access to the assets that custodian holds (compare Rev. Proc. 2025-31, § 6.02(3), with Rev. Proc. 2026-20, § 6.02(3)). And a contingent liquidity arrangement, permitted in both versions, now expressly excludes a transaction “that the trust treats as a borrowing of those digital assets for Federal income tax purposes” (Rev. Proc. 2026-20, § 6.02(12); see id. § 2.09(7)).

The transition rule is short. A trust that acts within six months after October 6, 2026 to conform, by amending its trust agreement, revising its procedures, or both, does not lose its classification by doing so. A trust that complied with the 2025 procedure may keep relying on it for the same six months. After that, “no trust may rely on the safe harbor in Rev. Proc. 2025-31” (Id. § 6.03). The window that followed the 2025 procedure was nine months (Id. § 2.09(8) (referring to “the 9-month grace period” under Rev. Proc. 2025-31)).

The limits matter more to most readers of this series than the conditions do.

The safe harbor is available only to a trust whose interests trade on a national securities exchange. A private trust holding digital assets for a family is outside it by the first condition. Section 7 then closes the door on reasoning by analogy: “No inferences should be drawn about whether similar consequences would result if actions taken by or on behalf of a trust fall outside the limited scope of this revenue procedure” (Id. § 7.01). The procedure also declines to say whether staking income is effectively connected income or unrelated business taxable income, and it says nothing on forks or airdrops (Id. § 7.02).

Nor does it need to reach private trusts. The power-to-vary test belongs to investment trusts, where many holders own undivided interests in a fixed pool. An ordinary discretionary trust is classified under the general rule, and its trustee is expected to exercise investment judgment. A family trustee who stakes is not, for that reason alone, running a business entity. Whether a particular arrangement crosses the line depends on its facts, and on questions the IRS has not answered.

Rev. Proc. 2026-20 should be read as evidence of how a careful regulator thinks about trust-held digital assets, and not as a rule for private trusts. Read that way, four points carry over.

Key control is ownership. The procedure reasons from key access to ownership in a single step: only the custodian can reach the keys, so only the custodian can sell or transfer. A creditor’s lawyer runs the same reasoning in reverse. If a settlor can move trust tokens from a phone, the trust instrument describes one arrangement and the blockchain describes another, and a court asked whether the settlor truly parted with the property will look at the second. Digital assets make retained control unusually easy to prove, because the ability to sign is the control. The procedure’s language also leaves an open question for split-key and multi-signature arrangements, in which no single party holds a complete key. It speaks only of a custodian with exclusive access, and it does not address those designs.

Related parties are the weak point. The procedure spends more words on the staking provider’s independence than on any other subject: unrelated to the trust and sponsor, serving other unrelated customers, paid on arm’s-length terms, bearing its own costs. A private trust whose validator is run by the settlor, or by a company the settlor owns, has none of that distance. Rewards that flow through an insider look like income the settlor still commands.

Staking has risks a trustee must own. Slashing, defined in the procedure as the forfeiture of staked units when a validator fails to follow the network’s rules, is a loss of trust principal (Id. § 2.05; see also id. § 2.04 (lockup and cooldown periods)). Lockup and cooldown periods can leave a trustee unable to fund a distribution. The safe harbor handles the first with an indemnity and the second with liquidity policies. A private trustee who stakes without having considered either has a fiduciary question to answer before any tax one.

The instrument should say so. The safe harbor assumes a trust agreement that authorises staking, and gives existing trusts six months to add the authority. A private trust deed drafted before staking existed may be silent. Silence invites a later argument about whether the trustee acted within its powers, and that argument is far easier to settle by amendment in a quiet year than in the middle of a dispute. Records matter for the same reason: the procedure measures its distribution deadline from the moment the trust “gains dominion and control” over rewards, the same moment the IRS has used to time the inclusion of staking rewards in income (Id. § 6.02(14); see Rev. Rul. 2023-14 (a cash-method taxpayer includes staking rewards in gross income in the taxable year in which the taxpayer gains dominion and control over them)). A trustee who cannot say when rewards arrived cannot say when they were taxable.

A year ago the question was whether a trust could stake at all without ceasing to be a trust. The answer is now yes, for one class of vehicle, on fourteen conditions. Nearly every one of those conditions keeps someone at a distance from the asset: the trustee from market timing, the sponsor from the validator, everyone but the custodian from the keys. The procedure does not bind a private trust. It does show, in a regulator’s own words, which facts make the Service comfortable that a trust is holding property and not operating it. A settlor who keeps the keys has the opposite facts.

This note is general information, not tax or legal advice. Rev. Proc. 2026-20 is a safe harbor for a specific class of exchange-traded trusts; how any other trust that holds or stakes digital assets is classified or taxed depends on its terms, its facts, and the law of the jurisdictions involved, and should be reviewed with qualified counsel.