USCIS AI translation errors in Pashto jeopardize Afghan asylum claims
US Citizenship and Immigration Services (USCIS) and its contractors relied on AI translation tools for Afghan refugee asylum claims, leading to critical errors in Pashto and Dari translations. These inaccuracies resulted in discrepancies that led to the denial of asylum claims.
AI translation tools should never be used to replace translators and interpreters and they should not be used in high-stakes situations.
Key facts
- What
- US Citizenship and Immigration Services (USCIS) and its contractors relied on AI translation tools for Afghan refugee asylum claims, leading to critical errors in Pashto and Dari translations.
- Incident date
- Apr 1, 2023
- Who
- US Citizenship and Immigration Services
- Failure mode
- Hallucination
- AI surface
- Chatbot
- Severity
- High
What happened
USCIS and government contractors utilized neural machine translation tools to process asylum applications for Afghan refugees. Because these tools are poorly trained for low-resource languages like Pashto and Dari, they introduced significant errors into official documentation. In at least one documented case, an automated tool changed "I" pronouns to "we," creating a contradiction that led a judge to reject an asylum bid.
What broke inside the model
- 01 · TriggerA user asks for a fact, a citation, or a figure.
- 02 · Model stepThe model writes a fluent, confident answer.
- 03 · Control gapNothing ties the claim back to a real source.
- 04 · FailureA fabricated fact ships as if it were verified.
- 05 · ConsequenceThe false claim reaches a customer, a court, or the public.
Confidence holds, and even spikes, as the claim detaches from any source.
The failure stemmed from the use of neural machine translation (NMT) on low-resource languages that lack comprehensive training data. The models failed to capture linguistic nuances and regional colloquialisms, leading to literal translations of idioms and pronoun errors. A lack of human supervision in high-stakes legal contexts allowed these technical inaccuracies to be weaponized as evidence of dishonesty.
What it cost
Sources
- PressAI translation jeopardizes Afghan asylum claimsrestofworld.org
- PressLost in AI translation: growing reliance on language apps jeopardizes some asylum applicationstheguardian.com
Cite this entry
https://failureindex.ai/failures/uscis-translation-errors-pashto-jeopardize-afghanAI Failure Index. "USCIS AI translation errors in Pashto jeopardize Afghan asylum claims" (FI-0514). Realm Labs. https://failureindex.ai/failures/uscis-translation-errors-pashto-jeopardize-afghan (indexed Jun 16, 2026).Data fields CC-BY 4.0, prose citation permitted. Incident ID FI-0514. Full dataset at /data.
Note from Realm Labs, the Index steward
How Realm would have caught this
- Prism
- OmniGuard
- AI Detection & Response (AIDR)
A runtime layer that watches the model's internal state can flag the moment a model commits to a claim it has no support for, and hold or reroute the response before it reaches a user. Realm reads those signals in real time rather than grading the transcript after the fact.