Thursday, 8 Oct 2026
|
Phase 3 of CBP's CAPE refund process for IEEPA tariffs opened on October 6, extending refunds to finally liquidated entries for importers with court orders. The money is real, but it only moves when the claim data matches the entry records. Of 27.2 million entries claimed so far, 6.1 million have failed CBP validation. For importers, customs brokers and the 3PLs that support them, the bottleneck is no longer the law. It is paperwork: finding the right entry summary, invoice and liquidation notice for every entry and making sure the numbers agree.
This post describes operational practice, not legal advice. Talk to your customs broker or trade counsel about eligibility for specific entries.
CBP's refund process for duties paid under the IEEPA tariffs has been rolling out in phases. According to a CBP court filing on September 15, Phase 3 opened October 6 and covers finally liquidated entries for importers that hold court orders (Sovereign Magazine).
The scale of the process so far:
The last row is the one operations teams should focus on. More than one in five claimed entries has failed validation. Each failure is money that has not come back yet, and each one needs someone to find out why.
CBP does not publish a single list of rejection causes, but the pattern in any high-volume claims process is familiar. Claims fail when the data submitted does not match the record CBP holds. Common sources of mismatch include:
Every one of these is a document problem. The information exists somewhere, in a broker's portal, an email attachment, a shared drive or a PDF scanned two years ago. The work is finding it and matching it.
For most of the last two years, tariff paperwork was a cost center. Teams filed entries, paid duties and archived the documents. Phase 3 changes the value of that archive. An entry summary that can be found and matched in minutes is a refund that gets filed this week. One that takes an afternoon of searching is a refund that waits.
For importers with thousands of entries, the difference is not marginal. Here is an illustrative example: if a team can locate and verify documents for 40 entries a day by hand, a backlog of 4,000 eligible entries takes 100 working days. If document processing cuts the per-entry effort enough to handle 400 a day, the same backlog clears in 10. The numbers will differ for every business. The structure will not.
Tariff policy is also still moving. Shippers are already asking what the next round of changes means for their landed costs, which we covered in our post on the questions shippers are asking about the US-China truce and the January 10 date. Clean records help with both the refund and whatever comes next.
The job is indexing and matching, which is exactly what agents do well:
1. Collect the documents. The agent gathers entry summaries, commercial invoices, liquidation notices and broker correspondence from email, portals and shared folders.
2. Extract the fields that matter. Entry number, importer of record, entry date, liquidation date and status, HTS lines and duty amounts by type.
3. Build an entry index. One record per entry with links to every supporting document, so nobody has to search again.
4. Check claims before submission. The agent compares the claim data against the extracted entry data and flags mismatches: a duty amount that does not reconcile, an entry in the wrong phase, a missing liquidation notice.
5. Work the rejections. For entries that already failed validation, the agent pulls the source documents, identifies the likely mismatch and routes it to the person who can correct and resubmit.
Classification data matters here too. Refund amounts depend on the duty paid per line, and lines are only as clean as the original classification. We looked at why that goes deeper than code lookup in our post on AI tariff classification beyond HS codes.
Importers own the claim. In practice, the documents are spread across the importer, the customs broker and sometimes a 3PL or forwarder. The cleanest setups give one team responsibility for the entry index and let agents handle the collection and matching across all three.
Brokers and 3PLs that support importers can offer this as a service. The same infrastructure supports ongoing compliance work, which we described in our overview of AI trade compliance agents, and the day-to-day customs correspondence covered in our guide to AI-powered compliance for customs clearance emails.
What is Phase 3 of the IEEPA tariff refund process? Phase 3 of CBP's CAPE process opened October 6, 2026 and covers finally liquidated entries for importers with court orders, according to a CBP court filing on September 15 (Sovereign Magazine).
How many refund claims have failed validation? Of 27.2 million entries claimed, 6.1 million have failed CBP validation (Sovereign Magazine).
How much has been refunded so far? About $100 billion of roughly $166 billion in IEEPA duties had been refunded by July 31, 2026 (Supply Chain Dive).
Why do refund claims get rejected? Usually because the claim data does not match CBP's entry record: wrong entry numbers, duty amounts that do not reconcile, incorrect liquidation status or importer of record mismatches.
Can AI help with tariff refund claims? Yes, on the document side. Agents can collect and index entry paperwork, extract key fields and flag mismatches before a claim is submitted. Eligibility decisions belong with your broker or counsel.
Phase 3 makes old entry paperwork worth money, but only if it can be found and matched. Index your entry summaries, invoices and liquidation notices, check every claim against the source documents before submitting and work failed validations by cause, so refunds land instead of stalling.
Debales deploys AI agents for customs document processing, entry indexing and trade compliance correspondence, so refund claims match the paperwork the first time. Book a demo.

Wednesday, 4 Nov 2026
Cargo theft spikes over Thanksgiving, and Q2 2026 losses more than doubled. A holiday-week check-call protocol for high-value loads, Tuesday to Cyber Monday.