Immigration law firms adopted generative AI and advanced document automation tools at a rapid pace. This shift changed how practices manage drafting, evidence preparation, and court filings. As managing partners and solo practitioners integrate these technologies, federal agencies are defining their own rules. The regulatory posture is moving quickly from both directions.
On one side, the agency evaluating your submissions is using artificial intelligence to screen files. On the other side, the court system overseeing deportations and appeals has established strict guidelines for how attorneys use generative AI. This two-sided regulatory environment means that firms must understand the exact policies of both the Department of Homeland Security and the Department of Justice.
Evaluating the landscape requires looking at documented policies rather than forum rumors. This guide breaks down the current posture of federal immigration agencies. We also examine what these rules mean for firms using popular practice management and document automation platforms like Docketwise and eImmigration. For a deeper look at the broader software market, refer to our guide on Legal AI for Immigration Firms: A Buyer's Guide.
USCIS is already using AI to screen what you file
The United States Citizenship and Immigration Services, which is a component of the Department of Homeland Security, has integrated artificial intelligence into its internal workflows. This is not a future plan. USCIS is actively using machine learning tools to handle the high volume of petitions and benefit applications submitted by law firms.
One of the core systems in use is the Evidence Classifier. This tool operates within the USCIS Electronic Immigration System. The system automatically tags, categorizes, and organizes submitted documents before a human officer reviews the case. According to federal agency reports, these machine learning tools help organize massive evidence files to streamline the adjudication process.
Internal Workflows and Final Authority
The Department of Homeland Security has stated its position on the boundaries of this technology. Human adjudicators retain the final decision-making authority. The agency maintains that AI tools do not grant or deny immigration benefits on their own. Instead, the software serves as an administrative screening and classification layer.
The Missing Outcome Data
Despite these assurances, the Department of Homeland Security has not published outcome-based metrics comparing AI-assisted processing to traditional manual reviews. This means there is no official public data showing whether AI-assisted screening impacts approval, denial, or Request for Evidence rates.
The lack of official metrics has led to concern among immigration advocates and attorneys. A 2026 Cozen O'Connor client alert reported a perceived rise in RFEs and denials. The firm observed that this increase coincided with the expanded use of artificial intelligence in the adjudication workflow. While this is outside bar commentary rather than an official government finding, it highlights a shifting reality for practitioners.
What This Means for Evidence Preparation
If automated systems are classifying and tagging your evidence before a human officer reads it, filing errors carry higher risks. Minor inconsistencies, mismatched dates, or incorrect exhibit labeling can cause the screening software to flag a file incorrectly. The structural integrity and internal consistency of your filings matter more than ever. Firms must ensure that automated drafting tools produce precise, highly organized evidence packages.
What DOJ's immigration courts now say about AI in your filings
While USCIS screens benefit applications on the administrative side, the Department of Justice handles litigation through the Executive Office for Immigration Review. EOIR runs the nation's immigration courts and the Board of Immigration Appeals. EOIR has taken a direct approach to regulating how attorneys use artificial intelligence.
On August 8, 2025, EOIR issued Policy Memorandum PM 25-40, titled "Use of Generative Artificial Intelligence in EOIR Proceedings." Signed by then-Acting Director Sirce E. Owen, this memorandum established the first nationwide policy on the use of generative AI in immigration court filings. You can read the full text of the policy on the Executive Office for Immigration Review website.
No Blanket Ban or Mandatory Disclosure
It is important to understand what PM 25-40 does not do. The memo does not impose a blanket ban on the use of generative AI by attorneys or self-represented respondents. It also does not require attorneys to automatically disclose to the agency whenever they use AI to write a filing.
Instead, the policy clarifies that the responsibility for accuracy remains entirely with the practitioner. Submitting hallucinated or erroneous content, such as invented case citations, fabricated quotations, or mischaracterized legal authority, violates existing rules of professional conduct.
Individual Judicial Discretion
While the national policy does not mandate disclosure, PM 25-40 leaves the door open for individual immigration judges. Individual judges retain the authority to issue their own standing orders. A judge may require specific disclosures or place strict limits on the use of generative AI in their specific courtroom. This makes it critical for practitioners to monitor local court rules. For more on this, read our analysis of how courts are writing their own AI rules, and they don't agree.
The Broader DOJ Context
The timing of PM 25-40 fits into a broader federal initiative. Office of Management and Budget Memorandum M-25-21 directed federal agencies to develop generative AI governance policies on a specific schedule. This timeline pointed toward broader, DOJ-wide guidance by the end of 2025. Because of this, the rules in PM 25-40 may eventually be superseded or extended by comprehensive DOJ policies. Practitioners should view the current memorandum as a baseline that could evolve.
The sanctions are already happening
The warnings in PM 25-40 are not theoretical. Courts have already issued severe sanctions to immigration attorneys who failed to verify their AI-generated filings. These cases show that judges will not tolerate artificial intelligence hallucinations in legal briefs.
The Ninth Circuit Asylum Brief Sanctions
In a case reported in June 2026, the Ninth Circuit Court of Appeals sanctioned immigration attorneys who submitted a brief in an asylum, withholding of removal, and Convention Against Torture matter. The brief contained multiple nonexistent legal cases and fabricated quotations. The attorneys relied on AI tools to draft the legal arguments but failed to verify the sources before filing. You can read the details in the bar report on the Ninth Circuit sanctions.
The One Thousand Dollar Federal Habeas Sanction
In a separate federal habeas case, another immigration attorney faced sanctions after filing an emergency motion for a stay of deportation. The attorney used the AI tool Claude to generate the motion. The software fabricated several legal quotations within the text.
The court sanctioned the attorney $1,000 for the error. However, the judge noted several mitigating factors that prevented a more severe penalty. The attorney quickly admitted to the mistake, immediately withdrew the flawed filing, and enrolled in an ethics-of-AI continuing legal education course.
These two distinct examples show that the defense of blaming the software does not work. You can read more about these issues in our guide on how courts are sanctioning lawyers for AI-hallucinated citations. The core lesson is clear. The mitigating factor that reduced the fine in the second case was the attorney owning the mistake immediately, not the fact that they used an advanced AI tool.
What this means for firms using Docketwise, eImmigration, or similar tools
Immigration firms must understand how these regulations apply to the specific software they use. The software market includes both practice management suites and specialized drafting tools. Understanding the differences between these technologies is the first step in maintaining legal AI immigration compliance.
Form Automation vs. Generative Drafting
There are two main categories of AI-assisted tools used in immigration law. Each carries a different level of risk under current guidelines.
First, there are form-automation platforms. Tools like Docketwise and eImmigration primarily help populate structured USCIS forms from client-intake data. The risk of hallucination in these tools is relatively low. This is because the data comes directly from client-supplied intake forms, and the target fields on the USCIS documents are highly structured. For a direct comparison of these two systems, see our detailed guide on Docketwise vs. eImmigration for Immigration Firms (2026).
Second, there are generative AI writing assistants. These tools are used to write custom briefs, RFE responses, or other persuasive narrative documents. These systems carry a much higher risk of hallucination. Because they generate novel prose and search for legal citations, they require rigorous human verification.
Features Within Your Existing Tools
Many firms use the built-in AI features of their primary case management software. Here is how the leading platforms handle these features:
- Docketwise: This dominant incumbent automates over 80 USCIS forms and supports client intake in 11 languages. Its AI writing assistant and document-data-extraction tools are offered through an integration with 8am IQ. These advanced features are an add-on layer rather than a native part of the base subscription plans.
- eImmigration: Owned by Cerenade, this full-suite case management system includes billing, a client portal, and intake. Its AI capabilities focus on document summarization and automated alerts. It uses a competitive, per-user pricing model where all standard features are included across plans.
Neither vendor has issued a formal public statement regarding PM 25-40 or the internal AI tools used by USCIS. These platforms are designed to help firms manage cases and automate documents, but they do not assume legal liability for the output. To learn more about software options, check our overview of the Best Legal AI for Immigration Firms (2026).
The Practical Compliance Rule
Regardless of the platform your firm chooses, the legal responsibility remains identical. Under both PM 25-40 and standard bar ethics rules, the attorney of record must personally verify every factual assertion, case citation, and cross-reference. You must check the work of your AI tool before filing.
Additionally, firms should proactively check for local standing orders. Because PM 25-40 allows individual immigration judges to set their own rules, you must confirm the specific requirements of your local court. You can find more details on general professional standards in our report on what the bar actually requires when you use AI.
FAQ
Does USCIS ban the use of AI-drafted immigration filings?
No. There is no blanket ban on the use of AI-drafted filings at USCIS. However, USCIS uses its own AI tools to screen, classify, and organize submitted evidence. The attorney remains responsible for ensuring all automated filings are completely accurate and internally consistent.
Does DOJ's EOIR require attorneys to disclose when they used AI to write a filing?
No, the agency-wide policy does not require automatic disclosure. Under Policy Memorandum PM 25-40, effective August 8, 2025, there is no nationwide disclosure mandate. However, individual immigration judges have the authority to issue local standing orders that require disclosure in their specific courtrooms.
Can an immigration attorney be sanctioned for AI-generated errors in a filing?
Yes. Attorneys are personally responsible for the accuracy of their filings. The Ninth Circuit sanctioned attorneys in June 2026 for a brief containing AI-hallucinated cases. Separately, a federal court issued a $1,000 sanction in a habeas case for Claude-generated fabricated citations, though the fine was mitigated because the attorney promptly admitted the error.
Does USCIS's use of AI make it more likely a petition gets an RFE or denial?
There is no official government data confirming a link. DHS has not published outcome metrics comparing AI-assisted screening to manual workflows. However, a 2026 client alert from Cozen O'Connor noted a perceived rise in RFEs and denials that coincided with the expansion of USCIS AI adjudication tools.
Are Docketwise and eImmigration affected by PM 25-40?
PM 25-40 regulates the conduct of practitioners in EOIR court proceedings, not specific software vendors. If your firm uses the AI drafting or summarization features of Docketwise or eImmigration to prepare documents for court, you must verify the output to ensure compliance with the memorandum.
The bottom line
The regulatory environment for immigration law is changing quickly. The agencies reviewing your filings are already using machine learning tools to classify and screen evidence. At the same time, the courts overseeing your cases have made it clear that you are personally liable for any errors produced by generative AI.
This means that immigration attorney AI responsibility is now a central part of practice management. Platforms like Docketwise and eImmigration offer powerful tools to help streamline your workflows, but they do not replace the need for careful legal review. The attorney of record remains the ultimate gatekeeper of accuracy.
As you integrate these systems, remember that the technology is an assistant, not a replacement for legal judgment. The guidelines under PM 25-40 represent a baseline, and broader DOJ policies may still emerge under the OMB M-25-21 directive. Staying compliant means verifying every citation, double-checking every document, and monitoring the specific rules of your local courts.