Why Immigration Professionals Should Think Twice Before Uploading Client Documents to ChatGPT
AI makes immigration paperwork easier. But when the information belongs to a vulnerable client, convenience shouldn't come at the expense of privacy.
Imagine you're a solo immigration document preparer. A client sends you photographs of their passport, birth certificate, and marriage certificate over WhatsApp. You need to extract their name, date of birth, passport number, and other details to prepare their immigration forms. You could manually enter everything into your software. But that takes time, especially when you're handling several clients.
Or you could upload the photographs to ChatGPT and ask:
"Extract all the information from these documents and organize it into a table."
Within seconds, you have everything you need. It's convenient, inexpensive, and saves a considerable amount of repetitive work.
But you've also just uploaded someone's highly sensitive identity documents to a third-party AI service.
And that person may have no idea.
I'm the founder of Fillvisa, a platform for preparing U.S. immigration forms. We work on software for individual applicants and immigration professionals, and I've spent a lot of time thinking about the privacy implications of this workflow.
I think we're approaching a serious problem.
The professionals I'm worried about aren't large law firms
Large law firms can invest in sophisticated case management software, secure document portals, IT professionals, and enterprise technology. But the immigration services industry also includes thousands of much smaller businesses.
Solo immigration attorneys. Independent document preparers. Small immigration consulting practices. Community organizations. Nonprofits helping low-income families. Some are one-person operations.
These professionals often handle extremely sensitive client information but operate on tight budgets. Consider the cost of specialized software. DocketWise, one of the established immigration case management platforms, currently starts at $90 per user per month on its monthly Basic plan. That's $1,080 per year for a single user. That may be reasonable for an established law firm. But for someone preparing a relatively small number of applications each month, it's another significant expense. And when you're trying to keep your services affordable for low-income families, every recurring expense matters. Now compare that with ChatGPT.
A free or relatively inexpensive AI tool can read documents, extract information, translate text, and organize data. I understand why someone would use it. Sometimes it's a lack of awareness about data privacy. Sometimes it's simply an economic necessity.
I don't think the solution is to shame these professionals for using AI. I think we need to provide accessible alternatives that don't require them to compromise their clients' privacy.
What happens to a passport after you upload it to an AI chatbot?
This is the question I wish more professionals would ask. A passport isn't an ordinary document. It can contain a person's full legal name, nationality, date of birth, photograph, passport number, and other identifying information.
Immigration case files can contain considerably more:
Current and previous addresses
Employment and financial information
Family relationships
Immigration history and status
Social Security and A-numbers
Medical information
Sensitive details about asylum claims or other legal matters
Now imagine uploading multiple documents belonging to the same client. Even if the immediate purpose is something simple - extracting a passport number - you may be transferring far more information than the task actually requires. When you upload a document to a cloud-based AI service, the information leaves your device and is processed by another organization.
What happens afterward depends on that provider's product, settings, contractual terms, and data-handling practices. For example, OpenAI's data controls documentation explains that conversations in personal ChatGPT workspaces may be used to improve its models unless the user disables the relevant setting. OpenAI's Business, Enterprise, and API offerings have different defaults: customer inputs and outputs are not used to train models by default. That's an important distinction.
It's inaccurate to say that everything uploaded to ChatGPT automatically becomes public or is used for training. OpenAI also states that it does not sell ChatGPT conversations or share them with advertisers. But turning off model training doesn't necessarily mean the content is never retained, processed by service providers, or subject to a lawful data request.
And a personal chatbot account shouldn't automatically be treated as an approved system for storing client documents. The question isn't whether every AI company is malicious. It's whether the professional uploading the documents understands the privacy implications and has taken appropriate steps to protect their client.
These privacy risks aren't hypothetical
There have already been several incidents demonstrating what can happen when sensitive information is uploaded to AI platforms or entrusted to online services.
Case 1: ChatGPT exposed information belonging to other users
In March 2023, OpenAI took ChatGPT offline after discovering a software bug that allowed some users to see titles from other users' conversations.
OpenAI also disclosed that payment-related information belonging to approximately 1.2% of ChatGPT Plus subscribers active during a particular nine-hour period might have been visible to other users.
The issue was fixed, and OpenAI published a detailed explanation.
Source: OpenAI - March 20 ChatGPT Outage: Here's What Happened
This incident didn't establish that immigration documents were exposed. But it demonstrated that even major AI platforms can experience security failures involving user information. Now imagine a similar mistake involving sensitive immigration case material.
Case 2: DeepSeek left chat histories exposed in an unsecured database
In January 2025, security researchers at Wiz discovered a publicly accessible database belonging to AI company DeepSeek. According to their investigation, the exposed database contained more than one million log entries, including chat histories, API keys, and internal system details.
The researchers reported that the database could be accessed without authentication. DeepSeek secured it after receiving their disclosure. There was no confirmation that unrelated malicious parties had accessed the information.
Source: Wiz Research - Exposed DeepSeek Database Leaking Chat History
The lesson here is straightforward. Even if a user trusts an AI model, the infrastructure surrounding that model can create additional risks.
Case 3: Samsung employees uploaded confidential information to ChatGPT
In 2023, Samsung employees reportedly uploaded sensitive internal source code to ChatGPT. Samsung subsequently restricted employees' use of generative AI tools because of concerns about confidential company information being transmitted to external services.
Source: Bloomberg - Samsung Bans Staff's AI Use After Spotting ChatGPT Data Leak
This is particularly relevant because the initial mistake wasn't a sophisticated cyberattack. It was employees trying to use AI to make their work easier.
That's precisely the kind of everyday situation I'm concerned about in immigration practices. A preparer doesn't need malicious intentions to create a privacy risk.
They might simply be trying to save ten minutes.
There's another reason immigration data deserves special protection: Government access
This is where the discussion becomes more serious.
For many businesses, a data leak primarily creates risks of fraud, identity theft, or financial loss.
Those risks exist for immigrants too.
But depending on a person's circumstances, the disclosure of addresses, identity information, or immigration-related records can carry additional consequences.
And we already have documented examples of private companies supplying personal information used for immigration enforcement.
Case 4: ICE obtained personal information through private data brokers
In 2022, researchers at Georgetown Law's Center on Privacy and Technology published American Dragnet: Data-Driven Deportation in the 21st Century.
After reviewing public records and government contracts, they documented how Immigration and Customs Enforcement (ICE) obtained access to extensive personal information through commercial data brokers and other databases.
According to the report, ICE had access to driver's license information, location data, and utility records covering large portions of the U.S. population.
The researchers estimated that ICE could locate approximately three-quarters of U.S. adults through utility-record data.
Source: Georgetown Law - American Dragnet
Think about what that means.
Someone might give their address to a utility company to obtain electricity or internet service.
They aren't providing that information to an immigration enforcement agency.
Yet commercial data arrangements can create pathways through which personal information becomes accessible to government investigators.
A separate 2022 investigation by The Guardian described ICE's contracts involving LexisNexis and data provided through Appriss, including information used to identify people in detention and monitor jail bookings.
ICE said access to such data was important for its law-enforcement operations. Privacy and immigrant-rights advocates argued that these arrangements could undermine protections intended to restrict certain forms of information sharing.
Case 5: DHS purchased commercially collected cellphone location data
In July 2022, the American Civil Liberties Union published records obtained through a Freedom of Information Act lawsuit.
The records showed that agencies within the Department of Homeland Security, including ICE and Customs and Border Protection, had purchased access to cellphone location information collected through commercial channels.
The data had been aggregated and sold by companies including Venntel and Babel Street.
Source: ACLU - New Records Detail DHS Purchase and Use of Cellphone Location Data
Again, the issue wasn't necessarily that individual users had knowingly shared their location with immigration authorities.
Instead, information collected for one purpose became available for another.
I think this should fundamentally change how we think about data collection in immigration technology.
To be clear, these investigations do not establish that ChatGPT, or any particular AI chatbot, supplied immigration clients' uploaded passports to ICE.
They document a different problem: how private companies can become sources of personal information for government enforcement.
I'm not equating AI providers with data brokers.
But these cases show why the destination of sensitive personal information, and the rules governing its subsequent disclosure, matter enormously.
Even immigration-specific software isn't immune
There's another incident worth acknowledging, especially because I'm building software in the same industry.
In October 2025, DocketWise identified a potential security issue involving credentials associated with a third-party partner repository.
Its investigation later confirmed that an unauthorized actor had used valid credentials to copy certain partner repositories involved in a data migration pipeline containing immigration law firm records.
The potentially affected information included passport numbers, Social Security numbers, financial information, medical information, and other personal details, depending on the individual.
A supplemental filing with the Maine Attorney General reported 143,480 affected individuals.
DocketWise said there was no evidence that the incident specifically targeted immigration firms or resulted in the public publication of personal information. It also reported remediation measures and notification of affected individuals.
Sources: DocketWise's own data incident disclosure and Maine Attorney General's supplemental filing.
This wasn't an AI chatbot incident, nor is it evidence that information was given to ICE.
But it's directly relevant to immigration software.
A specialized, paid platform isn't automatically secure just because it was designed for law firms.
Any service storing sensitive immigration records, including Fillvisa Plus, has a responsibility to protect that information.
I think we need more transparency throughout this industry about where data is stored, who can access it, how long it's retained, and what happens when something goes wrong.
Lawyers have professional confidentiality obligations, too
For licensed attorneys, this isn't just a technology question.
It's also a professional responsibility issue.
In July 2024, the American Bar Association published Formal Opinion 512, addressing lawyers' use of generative AI.
The opinion emphasizes existing duties relating to client confidentiality, competence, supervision, and communication.
It says lawyers must evaluate the risks of disclosure or unauthorized access before entering information about a client representation into a generative AI tool. Depending on how the tool works, informed client consent may be necessary.
Source: ABA Formal Opinion 512 - Generative Artificial Intelligence Tools
The State Bar of California has also advised lawyers not to put confidential client information into generative AI products that lack adequate confidentiality and security protections.
Source: California State Bar - Practical Guidance on Generative AI
This does not mean attorneys are categorically prohibited from using AI.
Appropriately vetted tools, contractual safeguards, limited data processing, and informed consent can make particular AI workflows acceptable.
The problem is treating consumer AI chatbots as interchangeable with confidential legal practice software.
There's also an important distinction between attorneys and non-attorney document preparers.
The specific attorney ethics rules don't automatically apply to every notary or document-preparation business. But anyone handling someone else's passport or immigration paperwork should take confidentiality seriously.
And document preparation shouldn't be confused with legal representation. In the United States, notaries and unaccredited preparers generally cannot provide immigration legal advice merely because they can help complete paperwork.
Why I believe free and open-source alternatives matter
After researching all of this, I keep coming back to one question:
What are smaller immigration professionals supposed to use instead?
It's easy to tell people not to upload documents to ChatGPT.
But if the alternatives are expensive, complicated, or inaccessible, that advice doesn't solve the underlying problem.
This is one of the motivations behind Fillvisa.
We're building three products, each serving a different need.
1. Fillvisa Free: Free, browser-based immigration forms
Fillvisa Free is designed for individuals and professionals who simply need to prepare immigration forms.
It's free, with no account required.
The form-filling application runs in the browser, and users' form data is stored locally on their device rather than in a Fillvisa cloud account.
There's no AI involved in extracting or processing the information.
Users can enter their details, complete supported USCIS forms, and generate filled PDFs.
For someone who just needs to prepare a form, I don't think they should have to pay for an expensive subscription or upload their passport into a generative AI chatbot.
Local storage isn't a complete security solution, of course. Shared devices, browser extensions, malware, and inadequate device security can still expose data.
But keeping sensitive form data on the user's device removes one potential category of risk: routinely transmitting that information to a remote form-storage service.
2. Fillvisa OS: Open-source immigration software
We also maintain Fillvisa OS, our open-source immigration form-filling software.
Anyone can inspect the source code, self-host it, and adapt it under the terms of its AGPL-3.0 license.
I believe this is particularly valuable for civic technologists, nonprofits, legal aid organizations, and developers working with underserved immigrant communities.
Instead of depending entirely on a proprietary vendor, organizations can examine how the software works and choose how to deploy it.
They can potentially build their own workflows around the forms without introducing generative AI services.
Open source doesn't automatically guarantee security or privacy. The deployment still needs proper maintenance, secure hosting where applicable, and responsible data handling.
But it provides transparency and control that aren't always available with proprietary tools.
And for organizations serving low-income immigrants, avoiding recurring software licensing fees can make a real difference.
3. Fillvisa Plus: Affordable collaboration for professionals
Of course, local form filling isn't enough for every professional.
Immigration firms also need client intake, document collection, collaboration between staff, and secure access to client records.
That's what we're building with Fillvisa Plus.
Plus provides a managed workspace with client immigration profiles, intake workflows, document management, permissions, e-signatures, and reusable data across multiple forms.
Our solo plan starts at $40/month.
I'm not trying to suggest that $40 is affordable for absolutely everyone. Or that pricing alone establishes security.
But I want professional immigration software to be accessible to smaller practices that may not need or be able to justify larger software subscriptions.
Unlike Fillvisa Free, Plus is a hosted application that handles information through a managed service.
That creates responsibilities for us as the vendor.
We need to be transparent about our data practices and continually improve the safeguards around sensitive client information.
Our users shouldn't simply have to trust us because we say our software is secure.
They should be able to understand what happens to the information they entrust to us.
A few practical rules I'd recommend
I don't think immigration professionals need to stop using AI altogether.
AI can be useful for researching general questions, drafting nonconfidential communications, brainstorming workflows, and explaining technical concepts.
But there's a difference between asking ChatGPT to explain the structure of Form I-130 and uploading an actual client's passport.
For small practices, I'd recommend starting with a few basic rules:
Don't upload identifiable client documents into an unvetted consumer AI chatbot. This includes passports, identity cards, medical records, and immigration case files.
Use dedicated intake and form-preparation workflows where possible. Choose tools based on their privacy and security protections, not simply their AI capabilities.
Understand where information goes. Review retention, deletion, subprocessors, model-training settings, and access controls before using a third-party service.
Obtain appropriate consent. Attorneys should follow the applicable professional rules; other preparers should be transparent about how client information is processed.
Minimize the information collected and shared. Don't upload an entire document when a less sensitive, narrowly scoped method will do.
Verify every extracted detail. A mistaken date of birth, passport number, or immigration history entry can have serious consequences in an immigration application.
A locally operated, non-generative document-processing tool may also be appropriate when it meets the user's security needs. The goal isn't to reject useful technology; it's to avoid unnecessary disclosure.
Privacy shouldn't be a luxury for immigrants
I want to end with something that goes beyond Fillvisa.
There's a lot of excitement about how AI will democratize software.
And in many ways, I share that excitement.
AI is making previously expensive capabilities available to people who couldn't afford them before.
But I worry about a future where people who can afford sophisticated systems get better privacy protections, while people on tight budgets end up using free consumer tools that weren't designed for handling sensitive client records.
Low-income immigrants shouldn't have to trade away privacy just to get help preparing paperwork.
And small immigration professionals shouldn't have to choose between expensive software and casually uploading client passports to whichever AI chatbot is available.
I believe there's room for another approach.
Free tools for people who need basic form preparation.
Open-source software for organizations that want control over their infrastructure.
And affordable, responsibly managed workspaces for professionals who need collaboration.
That's what I'm trying to build with Fillvisa.
The goal isn't to make immigration professionals stop using technology. It's to make privacy-conscious technology accessible enough that they don't have to take unnecessary risks in the first place.
Because behind every immigration form is a real person.
And the information they share to pursue a visa, a green card, or a better future deserves to be treated with care.