EU Finalises CBAM Reporting Guidance for 2026 Compliance Year

The Analysis

The dates that decide this file are not the dates in the headline. One is 1 January 2026, when the Carbon Border Adjustment Mechanism stopped being a reporting exercise and became a paying one. The second is 30 September 2027, when the annual declaration for the 2026 import year has to be filed and the matching certificates surrendered. The third is 1 February 2027, the day certificate sales open. What the Commission has now finalised sits between those three dates, and the two things it is described as settling are emissions-data thresholds and deadlines.

Strictly speaking, guidance settles neither of them in the way most people assume. Guidance is not legislation. It is the administration's account of how it intends to apply the legislation, and where the two disagree, the regulation wins. The instrument that creates the obligation is Regulation (EU) 2023/956, as amended for the definitive period by Regulation (EU) 2025/2083 of 8 October 2025. A guidance document can tell you how to calculate embedded emissions. It cannot create a duty the regulation does not contain, and it cannot remove one that it does. That distinction looks academic right up to the first time somebody tries to rely on a sentence in guidance during a verification.

So let me be plain about what I think this announcement is. It is not a relaxation. The phrase doing the rounds is reduced administrative burden, and I want to slow that down: what can be reduced is the number of keystrokes, not the identity of the person carrying the liability. That person is the authorised CBAM declarant, and the liability sits there whether the emissions figure came from the declarant's own installation, from a supplier's statement, or from a default value published by the Commission. Read the announcement as administrative relief and you will discover in 2027 that you read a sentence about paperwork as if it were a sentence about money.

Start from what changed on 1 January 2026, because a surprising amount of what is still being written describes the old world. Through the transitional period, from October 2023 to December 2025, declarants reported embedded emissions quarterly and paid nothing. There was no certificate, no surrender, no price. The point of those two years was to generate enough data to make the real instrument workable. From 1 January 2026 the rhythm changed and a financial obligation attached to it: the declaration is annual, the surrender is annual, and both fall on the same day, 30 September of the year following the import year.

Before any of that there is a gate. Importing CBAM goods into the customs territory from 1 January 2026 requires authorised CBAM declarant status. You apply for it to the national competent authority of the Member State where you are established, and it is what opens the CBAM registry, which is where the declaration is filed and the certificates are held. If goods in scope are being imported without that status, the difficulty is not a reporting difficulty. It is one that surfaces at the customs declaration, long before the CBAM deadline, and it does not get solved by better data.

The threshold is where most of the argument will happen. The definitive regime introduced a mass-based de minimis: an importer whose cumulative net mass of CBAM goods in the cement, iron and steel, fertilisers and aluminium sectors does not exceed 50 tonnes in a calendar year falls outside the obligations. Three features of that sentence matter more than the number itself. It is calculated per importer, not per supplier and not per customs declaration. It aggregates across every CN code in Annex I and across every indirect customs representative that importer uses. And hydrogen and electricity sit outside it entirely, so for those two the obligations run from the first import with no threshold at all.

Here is the exception that catches people. The exemption is written for importers, and under the definition in the regulation an indirect customs representative is not an importer, because an indirect representative acts in its own name but on behalf of another person. A representative filing for several clients therefore cannot apply the 50-tonne exemption to itself; the arithmetic has to be run against the person on whose behalf the declaration is lodged. If you are a trader who has always imported through an indirect representative and assumed the exemption belonged to the representative, that assumption needs checking against your own contract before it needs checking against anything else.

The second point about the threshold is that it is a state, not a status. Fifty tonnes is an annual cumulative figure, which means it can be crossed on any day of the year, and nobody announces the crossing to you. Take a plain case: 30 tonnes of aluminium plate arrives in January, 25 tonnes of steel coil in April. Neither shipment looks like a CBAM event on its own. By the end of April the cumulative figure is 55 tonnes and the obligation is live. There is no grace period and no retroactive reclassification of the January shipment into exempt. There is instead an importer who spent four months without authorised declarant status and without collecting installation data. The consequence is uncomfortable but simple: you have to prepare in January as if you will cross, because by the time you know you crossed, the data you needed was generated in a reporting period that has already closed.

Four clocks run on this file and they are not synchronised. The customs clock runs at the moment of the customs declaration, where the CBAM account number has become a required element for goods in scope and where a TARIC document code carries a de minimis claim on the face of the declaration. The declaration clock runs to 30 September 2027 for the 2026 import year. The cash clock starts on 1 February 2027, when sales open, and in practice it is the one that bites first. There is a fourth clock for goods arriving under inward processing: the obligation arises when the goods are transferred to free circulation, and the reporting period is set by the date of that transfer, not by the date of import. Anybody planning 2027 cash flow around a single date will get three of the four wrong.

From 2027 the cash clock is not annual either. Once imported CBAM goods exceed the annual threshold, the declarant has to buy, every calendar quarter, certificates corresponding to at least 50% of the emissions embedded in goods imported since the beginning of that year. That figure was lowered from 80%, and the direction of travel is the interesting part: the holding requirement was relaxed while the underlying liability was not touched. Pricing adds a second wrinkle. In 2026 the certificate price is published quarterly; from 2027 it moves to weekly publication. A weekly price sitting on top of a quarterly purchase obligation turns the timing of the purchase into a treasury decision with real variance attached, not a compliance checkbox.

Why now, and why through this instrument. The definitive period began on 1 January 2026 and the first declaration is not due until September 2027. That twenty-one month gap is unusual and it is deliberate: the reporting year has to close before the money can be calculated. What the guidance is doing is closing the questions that would otherwise be settled by whoever reached them first. Thresholds and deadlines are precisely the two categories where an unanswered question becomes an expensive one, because both of them shape behaviour during the reporting year rather than at the end of it. A threshold tells you whether to collect data at all. A deadline tells you when collection stops being possible. Neither can be fixed retrospectively.

There is one more asymmetry in this calendar, and it is the one I would put in front of a procurement team rather than a compliance team. The emissions data for the 2026 reporting year is generated at the installation during 2026, by a monitoring plan that had to be in place before the year began. It cannot be reconstructed afterwards. Approach a supplier in March 2027 and ask for verified installation data for 2026, and the honest answer in most cases will be that the monitoring was never set up to produce it, and that no verifier can certify a period they did not observe.

That is not a supplier being difficult. It is what accredited verification is. So the operative deadline on this file is not 30 September 2027. It is the point in 2026 at which your suppliers' monitoring plans and verifier engagements are still adjustable, and for most supply chains that point falls in the first half of the year. Miss it and the 2027 deadline turns into a formality you complete with default values rather than a calculation you complete with your own numbers.

Now the part of the announcement I think is being read backwards: the emissions-data thresholds. The choice facing most importers is between actual values, measured at the installation and verified, and default values published by the Commission. Default values are the fallback, not the discount. Under the published rules they carry mark-ups: 10% for 2026, 20% for 2027, 30% from 2028, with a separate 1% for fertilisers. Read those figures off the instrument yourself before relying on them, because the mechanism is what I want you to see. The mechanism is a price. You declare a higher emissions figure in exchange for not having to run a verified data chain, and the price rises every year. When the announcement says the step reduces administrative burden, what is being offered is a trade, not a gift: fewer verification steps, higher declared emissions.

Complex goods make the trade sharper. Where a product is made using precursors that are themselves CBAM goods, the precursor's embedded emissions are added into the final product, and actual precursor data can only be used if it comes from a verification report issued by an accredited verifier and covering the relevant reporting period. A supplier's letter is not that. A mill certificate is not that. Last year's plant spreadsheet is not that. Without the report, the precursor goes in at default value and the mark-up lands on top. Across the files I have seen, this is the single point that costs the most, because it is the one place where a company can do everything correctly at its own tier and still be pushed onto default values by somebody two tiers upstream.

Whether indirect emissions are in scope at all depends on the goods, not on your process. Cement, fertilisers and sintered ore bring indirect emissions into the calculation. Iron and steel, aluminium, hydrogen, and electricity when imported as goods, are direct emissions only. There is a second trap inside the same rule: the functional unit is not always a tonne of product. Cement is measured per tonne of clinker content and fertilisers per kilogram of nitrogen content, so a shipment that is half clinker and half additive does not report on its gross weight. Anyone who built their data collection around tonnes shipped has built it on the wrong denominator for at least two of the six sectors, and they will discover that when the numbers stop reconciling rather than when they start collecting.

Let me put a figure on it, stating every assumption. Assume an importer brings in 5,000 tonnes of an iron and steel CBAM good during 2026, comfortably above the 50-tonne line, holding authorised declarant status. Assume embedded emissions of 1.8 tonnes CO2e per tonne of product, which is 9,000 tonnes CO2e for the year. Assume a certificate price of 80 euros. Assume the free allocation adjustment is calculated on a CBAM benchmark multiplied by a factor that steps down each year, starting at 97.5% for 2026 and reaching zero in 2034, leaving 2.5% of embedded emissions as the 2026 surrender base. The obligation is then 9,000 x 2.5% = 225 certificates, or 18,000 euros. If the same importer cannot produce verified data and falls back to default values with the 2026 mark-up, the base becomes 9,000 x 1.10 = 9,900 tonnes CO2e, the surrender becomes 247.5 certificates, and the cost is 19,800 euros. The gap is 1,800 euros.

Hold that gap against the following year. Same shipment, same 9,000 tonnes, but the mark-up for 2027 is 20%, and let me assume the factor has stepped down so the surrender base is 10% rather than 2.5%. I am choosing 10% for the illustration and it should not go into anybody's budget; the actual annual steps are in the annex to the regulation and they are not mine to paraphrase. Verified data: 9,000 x 10% = 900 certificates, 72,000 euros. Default values: 9,000 x 1.20 = 10,800 tonnes CO2e, x 10% = 1,080 certificates, 86,400 euros. The gap is 14,400 euros on the same physical shipment, eight times what it was in 2026, purely because the factor moved and the mark-up moved with it.

That is the argument for building the data chain now, and it is an argument about money rather than about virtue. In 2026 the difference between a verified figure and a default figure is small enough that a reasonable company can decide to run a year on defaults while it learns the system. By the time the factor has stepped down a few notches, that same decision costs a multiple of what it cost in 2026. The trap is treating 2026 as the baseline. It is not a baseline, it is the shallow end of a ramp, and the ramp is in the annex. If you are modelling this in a spreadsheet, the term most likely to be missing from it is the cross-sectoral correction factor, which enters the free allocation adjustment alongside the benchmark and the annual factor.

One more boundary, and this one belongs in a contract rather than in a procedure. The authorised declarant carries legal responsibility for the accuracy of the declaration. It does not matter whether the underlying data came from the declarant, from a supplier, or from a third party. A contract between buyer and seller can shift who pays and who can be pursued. It cannot shift who the authority pursues. So if you are the declarant and your supplier will not commit to a verified report, you are not choosing between two data sources. You are looking at three consequences: absorb the mark-up, find a supply chain that can produce a verification report, or declare a figure under your own name that you cannot substantiate. The first two are cost questions. The third is a liability question, and it is the only one of the three that does not close when the year does.

On who sits outside the mechanism rather than inside it: goods originating in countries that apply the EU ETS, meaning Iceland, Norway and Liechtenstein, are not subject to CBAM, and neither are goods from Switzerland, whose system is fully linked. Those are origin-based exclusions, so the evidence burden falls on non-preferential origin, which is a different exercise from the preferential origin work most customs teams already run. There is separate treatment for Northern Ireland, for returned goods, for inward processing, and for groups that want to declare through a centralised representative. Each carries its own conditions and none of them can be inferred from the general rule. This is the part of CBAM where I would most resist generalising from one file to the next.

The mistakes I expect to see are not exotic. Guidance gets read as if it were the regulation. Reduced burden gets read as reduced liability. The 50 tonnes gets calculated per supplier or per entry instead of per importer per year, which is the version that will produce the largest number of surprised companies. An indirect representative assumes it can claim the exemption. Cash planning gets deferred to 2027 because that is when certificates go on sale, which ignores the quarterly holding requirement starting in the same year. And the 2026 surrender base gets carried forward as the long-run cost, which will make every budget from 2028 onwards wrong in the same direction.

So what goes on the calendar. Within the current quarter, settle in writing who the importer is on every CBAM line you file, and settle it with whoever files for you. Where an indirect representative is involved, the 50 tonnes has to be computed against the person on whose behalf the declaration is lodged, and that person needs to know their cumulative position before the year runs away from them. Within thirty days, stand up a cumulative net-mass register by importer and by calendar year across every Annex I CN code you touch, with a warning line at 40 tonnes. Twenty percent of headroom against a threshold that can be crossed on any working day is not much headroom. Within the same thirty days, confirm authorised declarant status or get the application in with the competent authority, because registry access is a precondition for everything downstream and it is not instantaneous.

By the end of the following quarter, run a coverage test on your data. For each CN code you import, can you name the installation, can you name the accredited verifier, and does a verification report exist covering the 2026 reporting period? Where the answer is no, you are on default values for that line and the mark-up is already in your numbers whether or not it has been calculated yet. Run the same test one tier up, for precursors, because that is where the answer turns out to be no most often, and because a gap at that tier contaminates the finished product rather than the component.

Then put three dates somewhere other than a news article: 1 February 2027, when sales open; the quarterly purchase obligation starting in 2027 at 50% of embedded emissions since the start of the year; and 30 September 2027 for declaration and surrender. Give each one an owner. I would give the middle one to treasury rather than to compliance, because from 2027 the certificate price publishes weekly and the choice of which week to buy in becomes a cash decision with variance on it.

I will stop on a boundary, which is where I usually stop. Everything above describes the mechanism as published and as it applies to a declarant in the general case. It is not an opinion on your file, and it does not tell you how a declaration is completed, which is not my desk. What is worth carrying away is narrower. The threshold is cumulative and can be crossed on any day. The three dates are three separate risks rather than one. And the cost of not having verified data rises every year whether or not anybody reminds you. Beyond that, do not let a sentence about administrative burden talk you into a decision about money.

↑ Back to the key points

— By Dr. Ingrid Voss