USCIS is using AI to process immigration filings

USCIS is using AI to process immigration filings

Here's what the government's own records say, and what it means for how we prepare cases.

Here's what the government's own records say, and what it means for how we prepare cases.

News

News

News

DHS publishes a public inventory of AI use across its agencies. The USCIS page was last updated July 15, 2026. A few entries stand out:

• PDF Intake for myUSCIS (deployed): USCIS discloses that a generative AI model, accessed through Amazon Bedrock, extracts data from PDF forms uploaded online and checks them against form-specific validation rules.

• Evidence tagging (deployed): A machine learning classifier tags evidence pages in ELIS so officers can jump straight to specific document types. A related tool separates and classifies supporting documents for I-539s, and now I-765, I-131 and I-824 filings and related RFE responses.

• Translation (pre-deployment): A generative AI tool to translate documents and support real-time interpretation for officers. USCIS has flagged it as high-impact.

• Internal AI assistant (pre-deployment): "PAiTH" would include a legal persona producing research summaries and INA/CFR citations for agency staff, with human review required before official use.

The update itself drew scrutiny. Reporting noted it arrived months after OMB's deadline, and that some "high-impact" designations across DHS were changed inconsistently. Separately, DHS has said it plans to use AI to help identify and redact sensitive content in FOIA responses. That DHS tool is still in pre-deployment and is classified as not high-impact, so it faces lighter risk-management requirements. If USCIS adopts it, it could affect A-file requests.

What this means in practice:

  1. Organization still matters. These tools tag and separate evidence for specific forms and systems. Clear exhibit labels, clean scans and one document per exhibit may help surface what officers need, and they're good practice regardless.

  2. Officer-side machine translation doesn't replace the certified English translations required under 8 CFR 103.2(b)(3).

  3. Consistency matters. Deployed identity-matching tools link records across USCIS systems using names, dates of birth and identifiers such as A-numbers and passport numbers.

  4. Precision matters. Some tools are deployed, some are pilots, some are still in development. Not every "USCIS uses AI" claim online reflects that.

The inventory repeatedly says humans make the decisions. But AI increasingly shapes what those humans see, and how fast. Prepare filings with both audiences in mind.

***Disclaimer: This post is for general informational purposes and does not constitute legal advice. Every case is fact specific. Please contact our office to discuss how this decision may affect your individual situation.***

DHS publishes a public inventory of AI use across its agencies. The USCIS page was last updated July 15, 2026. A few entries stand out:

• PDF Intake for myUSCIS (deployed): USCIS discloses that a generative AI model, accessed through Amazon Bedrock, extracts data from PDF forms uploaded online and checks them against form-specific validation rules.

• Evidence tagging (deployed): A machine learning classifier tags evidence pages in ELIS so officers can jump straight to specific document types. A related tool separates and classifies supporting documents for I-539s, and now I-765, I-131 and I-824 filings and related RFE responses.

• Translation (pre-deployment): A generative AI tool to translate documents and support real-time interpretation for officers. USCIS has flagged it as high-impact.

• Internal AI assistant (pre-deployment): "PAiTH" would include a legal persona producing research summaries and INA/CFR citations for agency staff, with human review required before official use.

The update itself drew scrutiny. Reporting noted it arrived months after OMB's deadline, and that some "high-impact" designations across DHS were changed inconsistently. Separately, DHS has said it plans to use AI to help identify and redact sensitive content in FOIA responses. That DHS tool is still in pre-deployment and is classified as not high-impact, so it faces lighter risk-management requirements. If USCIS adopts it, it could affect A-file requests.

What this means in practice:

  1. Organization still matters. These tools tag and separate evidence for specific forms and systems. Clear exhibit labels, clean scans and one document per exhibit may help surface what officers need, and they're good practice regardless.

  2. Officer-side machine translation doesn't replace the certified English translations required under 8 CFR 103.2(b)(3).

  3. Consistency matters. Deployed identity-matching tools link records across USCIS systems using names, dates of birth and identifiers such as A-numbers and passport numbers.

  4. Precision matters. Some tools are deployed, some are pilots, some are still in development. Not every "USCIS uses AI" claim online reflects that.

The inventory repeatedly says humans make the decisions. But AI increasingly shapes what those humans see, and how fast. Prepare filings with both audiences in mind.

***Disclaimer: This post is for general informational purposes and does not constitute legal advice. Every case is fact specific. Please contact our office to discuss how this decision may affect your individual situation.***

Newsletter

Stay informed with legal insights

Receive thoughtful articles, legal updates, and practical guidance from our team — delivered directly to your inbox.

Newsletter

Stay informed with legal insights

Receive thoughtful articles, legal updates, and practical guidance from our team — delivered directly to your inbox.