The court reporting profession stands at a dangerous crossroads. CAT software vendors are rapidly integrating AI features—promising improved realtime accuracy, automated suggestions, faster transcript production, and competitive advantages. Simultaneously, court reporters are using consumer AI tools like ChatGPT and Claude for editing, proofreading, and formatting assistance.
Both practices raise the same fundamental question: What’s actually happening to confidential transcript data?
If CAT software offers AI-assisted writing features, or if reporters are using external AI tools for any aspect of their work, they need to understand the confidentiality implications. The answer may shock many in the profession.
The Problem
Court reporters maintain one inviolable principle: what happens in the courtroom stays under their control. The stenographic record never leaves secure systems until it’s finalized and delivered to authorized parties. This isn’t just best practice—it’s a legal and ethical requirement of the profession.
AI changes everything.
CAT Software AI Features: Most AI features in CAT software don’t process data locally on the reporter’s computer. Instead, they send realtime feed data—the stenographic output as it’s being written—to external servers for processing. This means AI-assisted suggestions, corrections, and enhancements require transmitting live courtroom proceedings to cloud-based systems.
External AI Tools: Beyond CAT software, reporters are also using consumer AI platforms. When transcript text gets pasted into ChatGPT, files uploaded to Claude, or any cloud-based AI tool used for editing and formatting assistance, confidential data leaves the reporter’s control.
In both scenarios—integrated CAT features and external AI tools—confidential proceedings, sealed testimony, privileged communications, and classified information enter systems that reporters don’t manage, can’t audit, and cannot fully protect.
Once transcript data hits an external AI service, confidentiality obligations have been fundamentally breached.
Where Your Transcripts Go
With CAT Software AI Features:
When AI features are enabled in CAT software, here’s what typically happens:
Cloud Transmission: Stenographic output—the realtime text feed as strokes are written—gets transmitted over the internet to the software vendor’s AI servers or to third-party AI service providers. These servers could be located anywhere in the world.
Realtime Processing: The AI model analyzes the stenographic output in real time, providing suggestions, corrections, or automated enhancements to improve accuracy and readability.
Data Storage: The software vendor or their AI provider may store the data they process. This could be temporary (for session management) or permanent (for service improvement). The terms of service determine retention policies.
Training Data Risk: Unless explicitly prohibited by contract, some vendors may use customer data to improve their AI models. Proceedings could become part of training datasets, potentially accessible through future AI outputs.
Multiple Copies: Data transmitted to cloud servers exists in multiple locations—primary servers, backups, redundant systems, archived logs—creating numerous potential exposure points.
With External AI Tools:
The risks are similar but often more severe:
Direct Upload: When reporters paste transcript text into ChatGPT, upload documents to Claude, or use web-based AI tools, they’re directly transmitting confidential data to consumer AI platforms.
Broader Training Use: Consumer AI platforms often explicitly reserve the right to use inputs for training purposes (especially on free tiers), meaning transcript data could become permanently embedded in AI models accessible to millions of users.
Less Secure: Consumer AI tools typically have fewer security certifications and contractual protections than enterprise software, increasing breach risks.
The Legal Minefield
Consider these specific scenarios where AI-assisted court reporting creates serious legal exposure:
Sealed Proceedings
Cases under seal by court order are legally prohibited from public disclosure. Using CAT software AI features that transmit realtime data externally, or uploading any portion of a sealed transcript to ChatGPT or any other external AI service, may constitute direct violation of the court’s sealing order. Courts do not authorize software vendors, OpenAI, Anthropic, Google, or any third party to access sealed testimony.
Grand Jury Proceedings
Federal Rule of Criminal Procedure 6(e) mandates grand jury secrecy. Grand jury transcripts cannot be disclosed except under specific, limited circumstances. Transmitting this data to an AI service likely violates federal law. The penalties aren’t trivial.
Attorney-Client Privilege
Depositions routinely capture privileged communications. When you send that data to an external AI service, you’re potentially creating a waiver of privilege. Worse, if that AI company gets subpoenaed in unrelated litigation, your privileged communications could be exposed.
Classified Information
Federal proceedings involving classified material have extraordinarily strict handling requirements. Commercial AI services like ChatGPT, Claude, or Gemini do not meet—and cannot meet—the security standards required for classified information processing. Using these tools with classified transcripts violates national security laws. Period.
HIPAA-Protected Information
Medical-legal depositions contain protected health information. Unless your AI provider is a HIPAA-compliant business associate with a proper BAA in place, you’re likely violating HIPAA by transmitting this data to them.
The Professional Ethics Violation
The NCRA Code of Professional Ethics is unambiguous. Court reporters are required to protect confidential information and maintain the integrity of the record. Using CAT software AI features that transmit data externally, or using consumer AI tools with transcript content, appears to directly violate these fundamental obligations.
State licensing boards have similar confidentiality requirements. A single complaint about breach of confidentiality could result in license suspension or revocation. Careers could end because AI features were enabled in CAT software without understanding the implications, or because transcript text was pasted into ChatGPT for formatting assistance, thinking it was harmless.
What the Terms of Service Say
Court reporters should read the terms of service for both their CAT software’s AI features and any consumer AI platforms they use. They will likely discover:
CAT Software AI Features:
License grants allowing the vendor to process and potentially use transcript data
Disclaimers of responsibility for data breaches
Provisions about data retention and deletion policies
Language about “service improvement” which may include AI model training
Clauses indicating data may be processed by third-party AI providers
Terms stating reporters are responsible for obtaining necessary consents
Consumer AI Platforms (ChatGPT, Claude, etc.):
Broad license grants allowing the company to use input data
Provisions stating inputs may be used to train AI models (unless enterprise agreements prohibit this)
Disclaimers of responsibility for data breaches
Data retention policies that keep inputs even after apparent deletion
Permission to use data for “service improvement”
Clauses stating data may be processed in foreign jurisdictions
Language indicating users are responsible for ensuring they have the right to share input data
In both cases, these companies have carefully structured their terms to protect themselves while placing all liability on the reporter.
The Subpoena Risk
Here’s a scenario that should concern every reporter: An AI company like OpenAI or Anthropic gets subpoenaed in litigation. The opposing party knows that lawyers and court reporters have been using these tools, and they want to discover what was shared. The AI company is required to produce all data related to certain parties, topics, or cases.
Your confidential proceedings—which you thought were secure—suddenly become part of someone else’s discovery. The transcript excerpts you pasted in for “harmless” editing help are now evidence in litigation.
You have no control over this. You won’t even know it happened until it’s too late. The AI company has no attorney-client relationship with the parties in your proceeding and no obligation to assert privilege on their behalf.
The Data Breach
AI companies get hacked. It’s not a question of if, but when. Major technology companies with billion-dollar security budgets suffer data breaches regularly. When that happens to a company processing court transcripts, the consequences could be catastrophic:
Sealed testimony exposed publicly
Privileged communications disclosed to adversaries
Classified information compromised
Trade secrets revealed to competitors
Personal information of witnesses and parties stolen
Who’s liable? You are. You’re the one who violated confidentiality by transmitting data to an insecure external system.
“But My Vendor Says It’s Safe” / “But ChatGPT Says My Data Is Private”
Whether it’s CAT software vendors or consumer AI platforms, reporters should ask these specific questions:
For CAT Software Vendors:
Is all AI processing done locally on the reporter’s machine, or is data transmitted to external servers?
If data is transmitted externally, exactly where are those servers located?
Do you use third-party AI services (like OpenAI, Anthropic, etc.) to power your AI features?
What happens to transcript data after processing? Is it permanently deleted, or stored?
Is transcript data ever used to train or improve AI models?
Who has access to transcript data on your systems—employees, contractors, AI service providers?
What security certifications do you have (SOC 2, ISO 27001, FedRAMP, etc.)?
Have you ever been subpoenaed for customer data? What’s your policy?
Are you HIPAA-compliant with proper business associate agreements available?
What’s your notification policy if there’s a data breach affecting transcripts?
Will you indemnify reporters if there’s a confidentiality breach resulting from your AI features?
For Consumer AI Platforms: All of the above questions apply, plus: 12. Is my data used for training AI models? How do I definitively opt out? 13. For paid tiers, what additional protections exist compared to free tiers?
Get answers in writing. If vendors or platforms can’t or won’t answer these questions clearly, that silence is informative.
For most consumer AI platforms like ChatGPT free tier, the answer to “Is my data used for training?” is yes by default. While users can opt out through privacy settings, this must be actively done—the default setting allows data usage for training. Even paid tiers often retain broad rights to data unless enterprise agreements explicitly prohibit it.
The “Everyone Else Is Doing It” Fallacy
Some reporters say, “But everyone’s using the new AI features in their CAT software” or “Everyone’s using ChatGPT for editing and formatting. It’s so helpful!” Neither justification constitutes a defense.
When there’s a breach of confidentiality, “everyone else was violating their ethical obligations too” won’t protect a license, reputation, or liability exposure. The first major case involving AI use in court reporting and confidentiality breach will set precedent that affects the entire profession.
No one should want to be the test case.
Additionally, “everyone” is not doing it. Careful, ethical reporters who understand their confidentiality obligations are questioning these AI features and not using external AI tools with live case data. The fact that some reporters—or some software vendors—are being careless doesn’t justify carelessness from others.
What About “Offline” or Local AI?
CAT Software with Local AI:
Some CAT software vendors may claim their AI runs locally without transmitting data externally. If true, this addresses many confidentiality concerns. However, verification is essential:
Is the AI model truly running entirely on the reporter’s local machine?
Does it ever “phone home” with usage data, telemetry, or sample transcripts?
Are software updates pulling transcript data for quality assurance?
Can technical support access transcript data remotely?
Does the software require internet connectivity for AI features to function?
True local processing is fundamentally different from cloud-based AI and addresses most confidentiality concerns. But ironclad confirmation that nothing leaves the system is required.
External AI Tools:
Running large language models locally requires powerful hardware and significant technical expertise most reporters don’t have. True local AI tools exist but are rare and complex to implement.
For most reporters, the practical reality is: if using well-known AI tools like ChatGPT, Claude, Gemini, or any web-based service, data goes to the cloud. There’s no way around it.
The Critical Distinction:
The distinction between local and cloud-based processing is not a technicality—it’s the entire basis for whether confidentiality obligations can be maintained. Cloud-based processing fundamentally breaches confidentiality. Local processing, properly implemented, does not.
The Profession’s Future
The court reporting profession is facing a workforce crisis with a 90% failure rate in stenographic education. The profession desperately needs solutions.
AI is not the solution to the education crisis. Better pedagogy is. Methodologies that reduce cognitive load, eliminate unnecessary decision points, and allow students to develop flow-state writing habits address the actual problem.
AI might help with post-production editing and transcript refinement. But not at the cost of fundamental professional obligations to maintain confidentiality.
The profession cannot be saved by compromising the ethical principles that make court reporters trusted officers of the court. The moment court reporters become just another technology vendor processing data in the cloud, the profession loses the essential characteristic that makes it valuable: absolute trustworthiness with confidential information.
A Challenge to CAT Software Vendors, AI Companies, and the Profession
CAT Software Vendors should clearly disclose:
Whether AI features transmit data externally or process locally
If external processing is used, which third-party AI services are involved
Exactly what data leaves the reporter’s machine when AI features are enabled
Where that data goes and how long it’s retained
Whether customer transcript data is used for AI training
What security measures protect transcript data
What happens in the event of a subpoena or data breach
Whether HIPAA-compliant business associate agreements are available
AI Companies serving professional markets should clearly disclose:
Exactly what happens to user inputs—storage duration and access rights
Whether customer data is used for AI training, and how to definitively opt out
What security measures protect sensitive professional data
Policies regarding subpoenas, litigation holds, and data breaches
Whether HIPAA-compliant business associate agreements are available
Specific policies regarding confidential, privileged, or classified information
Court Reporting Agencies should establish clear policies about AI use and provide guidance to their reporters.
NCRA and State Associations should develop comprehensive ethical guidelines on AI use that protect both reporters and the integrity of the judicial record.
Court Reporters must take this seriously. The profession’s credibility depends on absolute trustworthiness with confidential information. This cannot be compromised for convenience or competitive advantage.
Conclusion
Court reporters who have handled sealed cases, classified proceedings, grand jury testimony, and depositions involving privileged communications have maintained confidentiality through one principle: maintaining absolute control over the record until it’s properly delivered to authorized recipients.
Both CAT software AI features that transmit data externally and the use of consumer AI tools with transcript data fundamentally compromise that principle. These technologies may offer convenience—better realtime accuracy, helpful suggestions, faster editing, improved formatting. But they come with legal and ethical costs that far outweigh any benefit.
Before enabling AI features in CAT software, before pasting transcript excerpts into ChatGPT, before uploading files to Claude for editing help, before using any cloud-based AI tool with case data, the question must be asked: Is the convenience or competitive advantage worth risking professional license, reputation, liability exposure, and the confidentiality rights of the parties?
Technology must serve ethical obligations, not undermine them. The convenience and competitive pressure of AI assistance are real and seductive. The confidentiality risks are catastrophic.