Why Public AI Is Now Dangerous for Lawyers

Bottom Line Up Front:

Your firm has a data leak. Regulators in Australia and the US have moved from caution to sanction. Using public tools like ChatGPT for client work is now a professional conduct breach. It can cost you your license. You cannot simply ban AI because your staff will use it anyway. To survive the 2025 regulatory shift, you must replace Public AI with Private AI.

The "Ostrich" Problem: You Can't Ban What You Can't See

Partners often believe they have solved the AI risk by banning ChatGPT.

This is a dangerous illusion.

According to the Menlo Security 2025 Report¹:

• 68% of employees use generative AI tools.
• 57% admit to pasting sensitive data into them.
• Most use personal accounts to bypass firm firewalls.

This is "Shadow AI." Your junior lawyers and paralegals face huge pressure to meet billing targets. They know AI can draft a document in seconds. If you do not provide a safe tool, they will use an unsafe one behind your back.

Ignorance is no longer a defense. If a junior lawyer pastes a refugee claim into ChatGPT, you are the one who loses your license.

The Regulatory Siege: It's Not Just a Warning

Regulators are no longer watching. They are acting.

1. The Australian Hammer: The Dayal Precedent

In late 2025, the Victorian Legal Services Board (VLSB+C) punished a principal lawyer, Mr Dayal, for AI misuse. The penalty was severe²:

• Stripped of rights to run a firm.
• Banned from handling trust money.
• Forced into supervised practice for two years.

The VLSB+C has drawn a red line. They state that lawyers "cannot safely enter confidential... information into public AI chatbots."³

2. The OMARA Trap

For Registered Migration Agents, the risk is even higher. Code of Conduct Section 35 forbids you from "allowing" disclosure to a third party.⁴

• The Breach: Pasting client data into free ChatGPT sends it to a US server. That is a disclosure.
• The Risk: OMARA has already suffered data breaches.⁵ They now apply heightened scrutiny to agents who leak data to foreign tech giants.

3. The US Warning Shot

In the US, courts are fining lawyers for AI "hallucinations" (fake cases). In Mata v. Avianca and Kohls v. Ellison, lawyers and experts were publicly shamed for filing AI content they didn't check.⁶⁷

This happens because of "Plausibility Bias." This means we trust computers because they sound confident, even when they are wrong.⁸

Public AI vs. Private AI: What is the Difference?

Most lawyers think "AI" is just one thing. It is not. You must know the difference between the App you use and the Model that powers it.

Public AI Apps (ChatGPT, Gemini, Claude)

These are consumer products. They connect you to a smart model, but the app records everything.

• Memory: The app saves your chat history to the cloud.
• Training: The provider uses your conversations to teach the AI.
• Privacy: Engineers can often view your chats to check for safety.

Private AI Workspaces (e.g. Grella, Phala Network, Red Pill)

These are secure workspaces. They connect to the exact same smart models, but they change the rules.

• Data Handling: Your data is readable only when the AI is working on it. It is locked and unreadable at all other times.
• No Training: Private AI apps sign contracts that ban model training.
• Ownership: You are the only one who can see your data. The provider is just a processor, not a storage unit.

The Solution: The "Private AI" Setup

You already use Clio, LEAP, Smokeball, or Actionstep to manage your practice because you trust their security. You need that same standard for your AI.

To fix the "Shadow AI" leak, you must give staff a sanctioned Private AI workspace.

The "Kill Switch": Zero Data Retention

To use AI legally in 2025, you need to change the software setup.

• Public AI: Uses "Retain and Train." Your client's story is stored. It is used to train the next model. It becomes part of the internet's permanent memory.⁹
• Private AI (The Compliance Standard): Uses Zero Data Retention.

Think of Public AI like writing on a blackboard. The words stay there, and others can learn from them.

Private AI is different. It is like writing with disappearing ink. The words are there only for the split second the computer is reading them to do the work. The moment the job is finished, the ink vanishes. Nothing is saved, so there is nothing to steal.

Top-tier Private AI saves your data in a Trusted Execution Environment (TEE). This creates a secure shelter for both the App and the powerful chips (GPUs) running the open-source models.

Imagine a Black Box made of magic steel. You put your client's file inside, the box does the work, and the answer comes out. But the box is locked from the inside. Not even the people who made the box or the company that owns the computer chips can peek in to see what is happening. Your data stays invisible to everyone.¹⁰¹¹

The Technical Reality: Where is your Data?

Do not be fooled by the "Sydney Server" myth.

• Data Residency: The hard drive is physically in Australia.
• Data Sovereignty: Australian Law controls the data.¹²

If you use a US-owned public AI (like standard ChatGPT), your data is subject to the US CLOUD Act. The US government can demand your client's data without a warrant. This bypasses other national courts. The right Private AI can ensure legal control stays in your country.¹³

Your Action Plan

Do not wait for a breach. Fix your process today:

1. Audit "Shadow AI": Survey your staff anonymously to find out what tools they use. Also, ask IT to scan for unauthorized email signups to see which AI sites employees have joined.
2. Update Engagement Letters: Add a clause that tells clients you use "Secure, Private AI (Zero Data Retention)." Transparency is your best defense.
3. Deploy the "Modern Stack": Pair your case management system (Clio/LEAP/Smokeball) with a Private AI workspace. Never use an AI tool unless they prove:
   • Data Ownership (You own all inputs and outputs).
   • Zero Data Retention (Contractual guarantee).
   • No Training on client data.
   • For a step further: ISO 27001 & SOC 2 Type II certification.

Technology speeds up your process. If your process is insecure, AI just speeds up your risk. Fix the model first.

---