ICE transferred more than 448,000 unaccompanied children to HHS across FY2019 through FY2023 — the DHS inspector general reported on 2024-08-20 that ICE has no automated system to track them and no formal process to tell HHS which ones missed a court hearing

What is happening?

Across fiscal years 2019 through 2023, Immigration and Customs Enforcement transferred more than 448,000 unaccompanied children to the Department of Health and Human Services, where the Office of Refugee Resettlement holds them and then releases most of them to a sponsor, usually a family member already in the United States.

On 2024-08-20 the Office of Inspector General of the Department of Homeland Security issued a management alert on that transfer chain. It found that ICE has no automated system for monitoring the location and status of these children after release, and that field offices instead rely on spreadsheets and electronic mail. The same audit counted more than 32,000 children who did not appear at scheduled immigration court hearings across the same fiscal years.

The audit was carried under project number 23-044-AUD-ICE and produced a final report numbered OIG-25-21 on 2025-03-25. A policy tracking record citing that final report adds two later snapshots: as of January 2025, more than 233,000 children had not yet been served a Notice to Appear, and as of October 2024, 43,000 children who had been served one did not appear. Those two figures rest on a single opened source and are graded accordingly in block 8.

The inspector general made two recommendations — build an automated system that manages court appearance and address information, and create a formal process that identifies children who did not appear and shares that information with HHS. ICE concurred with both and offered no implementation date, giving competing priorities and resource constraints as the reason.

Whose problem is this?

The transfer chain runs through two departments, and the segment after release belongs to neither of them in any system that can be queried. That is the shape of the problem rather than a description of any single decision inside it.

Where does this problem end?

The purpose row is the one that has to be read before the rest. A document about an absent tracking system can be read as an argument for more tracking of the same children and of the households that took them in, and that reading is a choice this document declines to make for the reader.

What is the state now, and what should it be?

Now

Should be

The target state is written in the two recommendations that ICE accepted — an automated system holding court appearance and address information, and a formal process that identifies children who did not appear and routes that fact to HHS. Both come from the auditing body and both were agreed to by the audited agency, so this is the standard the agencies themselves have on the record. What no opened source supplies is a date by which either is meant to exist, or a measure of how complete the tracking would have to be to count as done.

How big is it?

More than 448,000 children were transferred by ICE to HHS across FY2019 through FY2023. That is the cohort standing inside the untracked segment for at least part of that period, and it is the only figure in the opened sources that describes the cohort as a whole.

It is a floor and not a total. The phrase in the source is more than, and no opened source puts a ceiling on it. The annual figures that would extend the window forward are counted by a different agency on a different basis — the Office of Refugee Resettlement reports 118,938 referrals and 113,495 releases in FY2023, and 98,356 referrals and 99,381 releases in FY2024 — and referrals to ORR are not the same unit as transfers by ICE. Adding the two together would produce a number that no source states.

The population file for this document therefore records not-derivable rather than a range. The three audit sub-counts cannot be summed either, and block 8 says why.

Under what conditions does it arise?

1. The chain crosses a department boundary and the second half has no instrument. ICE holds the immigration case. HHS holds custody and then releases the child to a sponsor household. After that release the child is in a private home, the immigration case continues on its own schedule, and the record that would join those two facts lives in no system either department can query.

2. Manual methods produce local knowledge that does not aggregate. Spreadsheets and electronic mail at the level of a field office can answer a question about one child if someone knows which office to ask. They cannot answer a question about a cohort, and they cannot raise an alarm on their own.

3. Non-appearance is visible in one place and useful in another. The immigration court observes that a child did not appear. The agency holding the placement record is the one that could act on it. Between those two facts the inspector general found no accepted process, which is what the second recommendation was written to create.

4. An accepted recommendation carries no clock. ICE agreed that both fixes are needed and named competing priorities and resource constraints as the reason for not naming a date. Nothing in the audit process converts agreement into a deadline.

What has been tried?

Two of these attempts point at the audited gap and two point somewhere else. The 2025 memorandum and the 2025 interim final rule widened access to information about sponsors and about the households that received children; neither of them is the automated tracking system or the information sharing process the audit asked for, and no opened source reports that either recommendation was closed.

What was found?

These counts are not additive. The 233,000, the 43,000 and the 32,000 come from one audit project at three different reference dates, and they describe partly non-overlapping populations — one group had not been served a notice at all, another had been served one and did not appear. Adding them produces double counting, and no opened source performs that addition.

Why is it still unsolved?

Coordination failure — the transfer chain has two owners and the segment after release has none, and the audit that named this produced agreement rather than a date.

The first part is structural. A child moves from an enforcement agency to a social service agency and then into a private household, while an immigration case moves on a court calendar that neither agency controls. Each of the three legs has a custodian. The joins between them do not. The inspector general did not report that a system existed and failed; it reported that the system was never built, and that the work was being done in spreadsheets and electronic mail at the level of individual field offices.

The second part is that agreement is the end of the audit process, not the beginning of a project. ICE accepted both recommendations. Accepting a recommendation records that an agency does not dispute the finding, and it commits nobody to a date. The reason given was competing priorities and resource constraints, which is a statement about a budget rather than a plan, and no opened source reports a subsequent step on either recommendation.

The third part is that the same period saw a great deal of activity adjacent to the gap without closing it. A memorandum widened fingerprinting of adults in sponsor households, and a rule widened the sharing of immigration information about sponsors with law enforcement. Both of those move information about the adults who receive children. Neither of them is a system that tells the agency holding a placement record that a hearing was missed. The audit found the gap and the audited agency agreed with the finding, and neither of those steps sets a date by which anything must change.

What observation would mean it is solved?

Candidates — (a) ICE reports an automated system in operation for court appearance and address information and the inspector general closes that recommendation (b) a formal ICE and HHS information sharing process exists and both agencies describe it in a public document (c) the share of transferred children whose immigration court status can be retrieved from a single federal record rises toward all of them.

(a) alone is weaker than it looks. A recommendation closes when an auditor is satisfied that an agency did what it said it would do. A system can exist, hold addresses that were correct on the day of release, and be queried by nobody. Closure records an administrative fact about two agencies rather than an operational fact about a child.

(b) alone can be satisfied by a document. A process can be written down with no cadence, no named owner and no consequence for not following it. The audit asked for a formal process precisely because informal practice at field office level was invisible above that level, and a memorandum of understanding can reproduce that invisibility with better paperwork.

(c) alone counts retrievability, not contact. Knowing where a record says a child is located is not the same as anyone having reached that child, and the same completed record can serve either of the two purposes that block 3 declines to choose between. The three have to be read together, and any of them has to be read against what the record is then used for.

What is it connected to?

Fills with researchchild welfare data systems in the states that receive these placements, foster care and kinship placement oversight generally, immigration court backlog and its effect on hearing notice practice, the interoperability of federal case management systems across departments, and comparable handoff arrangements between immigration and child welfare authorities in other countries. Relation type and evidence grade were not confirmed in this round.

What these sources do not say

• Whether any child is unsafe. The audit measured three things — the absence of a system, the non-service of notices and the non-appearance at hearings. None of the three is a count of children whose whereabouts are unknown, and no source opened here supplies such a count or the method that would produce one. Public argument about this audit has used vocabulary the audit itself does not use; a published commentary examining that difference was not opened here, so the gap is recorded as a limitation rather than as a sourced counter-claim.
• Why the automated system does not exist. No source opened here says whether it was ever scoped, costed, requested in a budget, or declined. Competing priorities and resource constraints is the reason given for the absence of a date, and it is not an account of the absence of the system.
• Which agency is meant to own tracking after release. The recommendation names ICE as the builder and HHS as the recipient of shared information. No opened source states which agency carries the duty to know where a released child is, or whether any statute assigns it.
• The sample size behind the HHS inspector general findings. The number of case files reviewed for report OEI-07-21-00250 could not be confirmed, because the report page returned an error to every automated request.
• The effect of the 2025 policy changes. No source opened here measures whether widened information sharing about sponsors changed the willingness of sponsors to remain reachable, in either direction. This is not a case of a source reporting no effect; it is a case of no such evaluation being found.
• What the 2026 rule does. The Federal Register document published on 2026-06-26 could not be opened here, so its content, its effective date and whether it is in force are unconfirmed.
• Any statement from ICE beyond concurrence. The position of ICE on the record is agreement with both recommendations. No source opened here carries a separate statement from ICE disputing the finding, explaining the delay, or describing work under way.
• Any breakdown by place. None of these counts is broken down by state, by immigration court, or by the location of the care provider, in any source opened here.

See the evidence

No primary audit document was read in full. The federal oversight portal record of the management alert and the Government Accountability Office product page were opened directly and carry the FY2019 through FY2023 transfer count, the finding of no automated system, the non-appearance count, the two recommendations, the concurrence without a date, and the five of nine implementation status. The final report OIG-25-21 responded at its address but is served as an image-encoded file from which no text could be extracted, so every figure attributed to it here comes through a policy tracking record that cites it. That is why the January 2025 and October 2024 snapshots are graded medium while the management alert figures are graded high — the grade reflects what was opened here, not who did the measuring. The HHS inspector general report, the Federal Register document, the congressional hearing material and the commentary on how these figures have been characterized were all refused by their hosts to automated requests, and none of them was reached by any other route. Where opened sources overlap they agree: the transfer count, the absence of an automated system, the reliance on manual methods, the non-appearance count and the ICE response appear identically in the two sources opened for the management alert. Where the record is thin it is left thin rather than filled in — the reason the system does not exist, the assignment of ownership for post-release tracking, and the effect of the 2025 information sharing changes are all absent from every source opened here, and block 12 says so item by item. This is a Path A output, so observation_refs is empty and provenance_mode: press-derived.

People affected

Estimated range Not derivable

The reason and what is missing are listed under “What is missing” below