Withholding without a PE: six ways the UN tax convention drafts depart from US and OECD practice

Executive summary

Draft articles for the UN Framework Convention on International Tax Cooperation and both early protocols were published on July 20th and 21st, 2026, and taken up at the Intergovernmental Negotiating Committee's fifth session in New York, August 3rd-13th, 2026. Revised texts are due before the Nairobi session, November 30th-December 11th, 2026.

Why taxpayers should care: the drafts are not binding law today, but they may become the blueprint for source-country withholding rules, and the nearer-term risk is domestic-law adoption - countries citing the drafts to justify or widen taxes on cross-border services, digital services and insurance premiums.

What the drafts add is multilateral backing for that charge, imposed with no permanent establishment (PE) requirement. For United States (US) groups the central risk is asymmetric. The US walked out of the process in February 2025, but source countries can still impose uncapped domestic withholding on US providers, while a competitor resident in a country that has signed alongside that source country receives the protocol's rate ceiling, net-basis election and dispute protections.

For tax departments the work starts before any of this is settled: the exposure has to be found in the payment data before it can be priced, contracted around or provisioned for.

Key risks for US taxpayers

• Gross-basis exposure: withholding on active service income with no presence test, which on a low-margin engagement can exceed the profit on the transaction.
• No protocol relief: no access to the rate ceiling, net-basis election or arbitration, because those run only between signing States.
• Sourcing conflict: the same fee can be foreign-source where it is taxed and US-source at home, which can leave the withholding tax uncreditable.
• Credit leakage: the section 904 limitation and the section 960(d) haircut mean that in the net CFC tested income basket, where no carryover is allowed, unused credits are a permanent cost rather than a timing difference.
• Contract and pricing exposure: gross-up clauses, and transfer-pricing consequences where the cost lands on a service hub or principal company.

Why this reaches US groups

The United States is not at the table. On February 3, 2025, its acting representative to the Economic and Social Council, Jonathan Shrier, told the Committee's organizational session that the Convention's goals were inconsistent with US priorities and amounted to unwelcome overreach, and that the US would take no further part in the process.

The drafts depart from the OECD Model Convention and from US treaty policy on where service income should be taxed. Joining is optional; the terms are not. The rate ceiling, still blank in the drafts, and the net-basis election - the right to be taxed on profit rather than on gross income - are set in the protocol rather than bargained treaty by treaty, and the services protocol bars reservations outright - Article 19 provides that no reservations may be made to it - so a country accedes whole or not at all.

For a US business, it is worth asking whether staying out is worse than being in. Articles 5 to 7 of the services protocol apply only between two countries that have both joined, so the protocol never reaches a US provider's transaction directly. The charge arrives by another route: the paying country's own law, in many cases a withholding tax it enacted long before these drafts existed. What the protocol adds is international backing for it. The rate ceiling, the net-basis election and the dispute process all run between signing countries, so a competitor based in a signing country gets the capped version of the charge and a US provider gets the uncapped one.

Short of the source country's own law, only a US treaty can limit the charge, and coverage is uneven across the jurisdictions where this is most likely to bite. The US has no income tax treaty with Brazil, Nigeria or Kenya, so in those markets nothing above domestic law constrains the rate at all. It does have treaties with India, Indonesia and South Africa - but coverage is not the same as protection: the US-India treaty's fees-for-included-services article permits source taxation at 10 to 15%, so a treaty may cap the charge rather than block it.

Existing treaties are not left alone either. Convention Article 21(2) preserves rights under agreements already in force, but Article 21(3) obliges Parties to work toward alignment, to renegotiate where alignment requires it, and to negotiate in good faith when asked; Article 21(4) adds a duty to report to the Conference of the Parties. There is no deadline and no expiry, so the pressure persists. The OECD's 2016 Multilateral Instrument (MLI) amended much of the treaty network automatically once both partners notified a treaty. Article 21 works the other way: it sends countries back to the table pair by pair.

Who gets targeted

The drafts name no countries. The targeting is structural, and three filters do the work.

Payment type: three kinds of payments are covered - ordinary service fees, digital services delivered with little human involvement, and insurance and reinsurance premiums (services protocol Articles 5, 6 and 7). The protocol applies to taxes on service income however they are imposed or described, naming digital services taxes and equalization levies among them. Consumption taxes such as value added tax (VAT) and goods and services tax (GST) are left out. Whether international shipping and air transport are also out is unresolved: Article 8 would exclude income from international traffic, but the whole article sits in square brackets.

That matters because Pillar One would have required countries to drop their digital services taxes. This protocol keeps them, gives the idea behind them treaty backing, and in exchange caps what one signing country can charge another. A US business gets the legitimized tax without the cap.

Taxpayer: this filter barely filters at all - no revenue threshold, no PE, no physical presence. Articles 5 to 7 apply to any business based in a signing country that receives a covered payment, so a mid-market consultancy, a regional insurer and a hyperscaler are in scope on identical terms. Note which way that cuts: being in scope is what earns the cap, so a provider based outside the signing group is spared the ceiling, not the charge.

Jurisdiction: Article 20 of the Framework Convention requires a State to opt in to each protocol. A State joins the Convention first, then accedes to each protocol separately. The rules reach only pairs of States that have joined the same instrument.

The countries driving the text are the Africa Group and the wider Group of 77 (G-77), and the payments they care about are the ones flowing out to service providers based in wealthier countries.

Treaty practice is a better predictor than rhetoric, and it has been thin. The UN Model's service-fee provisions have reached only a handful of developed-developing treaties under Article 12A, introduced in 2017; two under Article 12B, introduced in 2021; and none at all under the broader Article 12AA, which replaced Articles 12A and 14 in March 2025 and is too new to have generated treaty practice either way. That may mean the demand is narrower than the rhetoric suggests, or it may mean the demand exists and cannot get past the partner whose providers would pay. If the second explanation is right, the multilateral route is the way past that blockage: a country signs once, the rule applies with every other signer, and no partner holds a veto.

Six ways the drafts diverge

The benchmark here is the OECD Model Convention and the MLI, not the two pillars. That is the layer these drafts operate on, and it is where US treaty policy and OECD guidance broadly agree. The UN drafts break with both on six points, mostly by taking the approach the UN Model has offered in Articles 12A and 12B for years, which lets the paying country tax, and moving it from a treaty-by-treaty option into a single agreement.

1. Nexus

Taxing rights come from domestic law, and treaties restrain them - on business profits an OECD-model treaty confines the source country to what is attributable to a PE. Services protocol Articles 5 to 7 remove that confinement between signing States: what creates the taxing right is the payment itself, so a provider can owe tax where it has no office, no staff and no equipment.

2. Sourcing

Article 5 covers services delivered by people - consulting, engineering, management - and asks first where the work was physically done, then where the customer is, then where the payer is or whether the payer deducts the cost. Article 6 covers services delivered with almost no human involvement, such as cloud hosting and online advertising, and never asks where the work was done: it starts with the customer, then looks to end users and user data there. The US Treasury has proposed the opposite approach on exactly that ground: regulations issued in January 2025 alongside T.D. 10022 would source cloud income by a three-factor formula keyed to the provider's own intangible property, personnel and tangible property (Prop. Reg. section 1.861-19(d)). Those sourcing rules are proposed, not final. If they are adopted, the same fee could be foreign-source where it is taxed and US-source at home - and since the foreign tax credit (FTC) limitation is computed on foreign-source income, a US-source characterization can leave that withholding tax uncreditable. The boundary between the two articles also moves: the services protocol never defines “service” and Article 3(2) defers to the tax law of the country applying it, so the same cloud payment can be a royalty in one counterparty's hands and an Article 5 or Article 6 fee in another's.

3. Tax base

This is the change with the largest cash effect. Services protocol Article 9 offers net-basis treatment, but by election rather than default: the election sits in Articles 5(4)(b), 6(6)(b) and 7(4)(b), and can be made only through whatever administrative procedure the source country provides. A provider with no physical presence that elects in is not taxed on its actual net profit from the engagement but on a reasonable allocation of profits, weighing gross revenues in that State against its total revenues from the activity.

Taxing the full payment rather than the profit is not foreign to US law. The US does it too, taking 30% off the top of many payments to foreign recipients (sections 871(a) and 881). What is foreign is applying it to active services income with no presence test, and the effect shows up fast: for a low-margin service business, a mid-single-digit rate on the full invoice can exceed the entire profit on a given transaction.

Worked example: assume a US provider bills $10 million of engineering services into a source country at a 15% pretax margin, so $1.5 million of profit. A 7.5% gross-basis withholding tax on the full invoice is $750,000, or half the profit on the engagement. If $400,000 of that is creditable in the US, the remaining $350,000 is an absolute cost and the engagement's margin falls from 15% to roughly 11.5% before any gross-up negotiation. A competitor resident in a signing jurisdiction, charged at a capped rate and able to elect net-basis treatment, is not in the same position.

ASC 740 consequences: the accounting follows the uncertainty. Where creditability under sections 901 and 903 is unresolved, the position may require measurement as an uncertain tax position; where credits arise in the net CFC tested income basket and cannot be carried over, a valuation allowance analysis follows; and where gross-up clauses shift the cost between the parties, the effective tax rate moves. These are provision-level questions, not only cash-tax questions, and they arrive well before any competent-authority case is filed.

4. Scope

Nothing narrows the net from there. There is no revenue threshold and no partial opt-out, so a country takes the protocol whole and every covered payment to a provider in another signing country falls inside it.

One qualifier runs the other way, and it is easy to miss. Article 1(3) switches the protocol off where the income is lightly taxed in the provider's home State - either because the statutory rate is at or below a threshold still left blank, or because a special exemption, exclusion or reduction tied to that income or entity brings the tax below what the threshold rate would have produced. What this narrows is not the charge but the protection. A US group delivering services through a low-taxed hub in a signing State can lose the cap, the net-basis election and the dispute route precisely because of the rate the hub was chosen for.

5. Relief

Between signing States, services protocol Article 10 makes the home country give credit for the tax the source country took, but only up to what the home country would have charged on that same income. Anything above that is not refunded, and the article prescribes the credit method rather than exemption.

US law further restricts relief where the increased withholding falls on a US group's CFCs. The section 904 limitation, combined with the 10% haircut on the deemed paid credit under section 960(d) - reduced from 20% by Pub. L. No. 119-21 for tax years beginning after December 31, 2025 - means a high gross-basis source tax may not be fully offset against US tax. In the net CFC tested income basket, where no carryover is allowed, the unused credit is a permanent cost rather than a timing difference. Whether the levy is creditable at all, under sections 901 and 903, is a separate open question.

6. Disputes and information

The dispute protocol is the strongest part of the package. Mandatory arbitration triggers at three years - or any longer period the competent authorities have agreed - before a three-member panel, on the request of any participating authority. The panel decides through an independent-opinion process unless the competent authorities agree it must instead choose one of their proposed resolutions without modification. The MLI's arbitration part, by contrast, is opt-in, runs two years and defaults to last-best-offer arbitration, in which the panel must adopt one side's figure rather than write its own.

Which procedure governs where a treaty already provides one is unsettled. Article V.2 of the dispute protocol offers two opposite answers, both still in square brackets: under Option A the mechanism in the existing instrument continues to apply, and the protocol's reaches only where that instrument has no substantially similar mechanism; under Option B the protocol's mechanism supersedes it unless a Party has entered a reservation. Reservations are made at ratification, acceptance, approval or accession under Article V.1, so the route a taxpayer gets turns on the option chosen and on what its counterparty reserved.

Information exchange is the one place the drafts may fall below the current standard. Article 11 of the Framework Convention, the information-exchange article, is bracketed in full, and its first paragraph carries two competing versions of the standard itself: information that “may be” relevant, or that “is foreseeably” relevant. The services protocol's own exchange article already uses the settled formulation, so the divergence is at Convention level. Foreseeable relevance is the settled formulation in OECD Model Article 26, and it is what stops a request from becoming a fishing expedition. A looser standard would not reach a US group through the IRS, since the US is not a party. It would reach it sideways, through subsidiaries, customers and banks in signing countries, which could be asked for more information about more taxpayers.

If adoption is broad

Assume the caps land near the fees-for-technical-services rates already common in developing-country treaties, and assume ratification across the G-77 with existing domestic charges left in place. Seven consequences follow.

Pricing moves first: a tax on the full invoice comes off the top, so someone has to absorb it - the provider raises the price, the client agrees to cover it, or the provider eats it. Where none of those work, the fourth outcome is no contract at all. Parties will fight over gross-up clauses long before the first competent-authority case is filed. A gross-up also feeds the charge it is meant to cover, since the tax applies to the enlarged gross amount.

Labels matter more than structures: with no PE condition, interposing a service company adds little, and what is worth planning around is how the payment is characterized. The same contract can be a royalty in one country and a service fee in the next, which changes the article, the ceiling, and sometimes whether the protocol applies at all.

Transfer pricing absorbs the cost: an unrelieved withholding charge has to sit somewhere in the group. Whether it stays with the entity that billed, is pushed into the cost base of an intercompany service charge, or is borne by a principal company, the answer has to be defensible under the arm's-length standard and consistent with the intercompany agreements. Gross-ups negotiated with third parties tend to pull intercompany pricing with them.

Operating models come under review: regional service hubs and principal-company structures were built on the premise that a source country could not tax without a PE. Where that premise no longer holds between signing States, hub locations, contracting entities and delivery models are all in play - and a hub resident in a signing jurisdiction may be better placed than one outside it, which is a new consideration rather than a familiar one. But only if the hub is taxed at a normal rate at home: under Article 1(3), a lightly taxed hub in a signing State loses the protocol's protections while remaining exposed to the source country's charge, which is the worst of both positions.

Digital revenue is the hardest to characterize: software-as-a-service, cloud, advertising, marketplace and AI-delivered services sit precisely where Article 6, Article 5 and the royalty definition overlap, and the answer can vary by counterparty. Contract drafting, invoicing detail and revenue-system data all become tax-relevant inputs.

The two-track world becomes durable: the OECD's Pillar Two side-by-side package, agreed January 5, 2026, allows US-parented groups to elect out of the income inclusion rule and the undertaxed profits rule while leaving domestic top-up taxes in place, and Pillar One Amount A remains unimplemented. Depending on final implementation, a US group can sit outside Pillar Two's charging provisions, outside a convention its government rejects, and squarely inside a source-State withholding charge that convention has normalized. Nonparticipation is not insulation.

The reciprocity argument runs out: the US has traded away source taxation throughout its network, and the 2016 US Model zeroes royalty withholding and has no fees-for-technical-services article. If a large bloc adopts the opposite default by multilateral instrument, that position has to be defended rather than assumed. Treasury's remaining levers are bilateral negotiation, creditability determinations and trade measures.

Bottom line

The drafts are further along than their blanks suggest. The architecture is settled; only the numbers and the reach are not.

Two things remain open: the rate, which sets the ceiling between signers and anchors expectations elsewhere, and adoption, which decides how widely the pattern spreads. The bilateral record argues for patience on the second. But low adoption is not low exposure. The charge that reaches a US provider is domestic law the protocol does not cap, much of it already on the books, and what accession changes is the politics of removing it: a country that has endorsed the charge multilaterally is a harder counterparty to talk out of it.

Watch list before Nairobi:

• Whether the caps in services protocol Articles 5 to 7 land as single rates or ranges; and not only where the Article 1(3) low-taxation threshold is set, but how its second prong is drafted, since that is what decides whether a preferential regime costs a provider the protocol's protections.
• Whether Article V.2 of the dispute protocol resolves to Option A or Option B.
• Whether the information-exchange standard remains “foreseeably relevant.”
• How broadly the protocols are ratified, and whether acceding States trim their domestic service charges toward the caps or leave them in place.
• Whether those domestic charges qualify for credit at all under sections 901 and 903.
• How major G-77 jurisdictions draft their implementing legislation.
• How gross-ups are treated for transfer pricing and under ASC 740.

Recommended next steps:

• Map service, digital-service and insurance payment flows into the jurisdictions most likely to accede.
• Model gross-basis withholding under several rate-cap scenarios, including the case where no cap applies.
• Review treaty coverage, gross-up clauses, withholding assumptions and foreign tax credit positions for the highest-exposure corridors.
• Prepare a creditability assessment under sections 901 and 903 for any material source-country levy reaching CFC tested income.

The broader point is control design. Exposure of this kind has to be monitored at the entity level, the payment level and the provision level, and the inputs required to do it - treaty status, domestic withholding rules, payment characterization, contract terms and general-ledger detail - typically sit in several different systems.

This article is general commentary on draft instruments that remain subject to negotiation and does not constitute tax, legal or accounting advice. Article references reflect the July 2026 drafts and may change. No conclusions should be drawn for particular facts without advice on those facts.

Hagit OrenFounder and CEO, TaxRay

Built and ran the global tax function at monday.com through the 2021 IPO, after a decade advising US multinationals on cross‑border structure as a Tax Partner at Herzog Fox & Neeman and Tax Director at Deloitte Israel.

Katie TuckerHead of Tax, TaxRay

Ran the tax department of a listed multinational operating in 17 countries, owning worldwide transfer pricing strategy, ASC 740 and SEC disclosure.