Shall we Chat?
AI, which has been around for more than 30 years, but changed significantly in 2025 when it began moving beyond giving answers to performing actions. Those actions often leave a trail of detritus behind.
How so? Let’s take a look at Anthropic’s warning about its’ AI product Claude Cowork. Anthropic cautions that Cowork can overwrite and permanently delete local files. Imagine having the only draft of a book you are writing suddenly vanish from your electronic files. Even worse, you may not be able to later reconstruct what happened. Anthropic states that Cowork activity is not captured in Audit Logs, the Compliance API, or Data Exports.
The dangers go further. Depending on the access you have given it, Cowork could read a malicious email you received and act upon it, such as downloading malware to your computer or transferring money from your bank account to a scammer. That’s not good.
We have all had to learn new terms to describe catastrophic outcomes from AI usage.
AI Terms You Need to Know
False Completion
“False completion” means the system gives you a false report that it has completed a task. Suppose you instruct the AI agent to prepare a spreadsheet that compiles information from source documents, but the AI agent in reality skips every other page. The resulting spreadsheet would be wrong.
Hallucination
“Hallucination” is a made-up result, such as citing a law that does not exist. Plenty of lawyers have been sanctioned for including hallucinated law in their pleadings and briefs. It’s clear that AI can produce false legal information.
Prompt Injection
“Prompt injection” is when an attacker uses inputs that manipulate the AI agent to ignore its original developer instructions and execute malicious actions. This is the danger that makes headlines when it goes wrong.
There are a lot of AI agents, each with their own strengths and vulnerabilities. Of the more popular ones, ChatGPT is often recommended for its everyday answers, reasoning, deep research and image generation, Claude gets the nod for writing and coding, Gemini has good reviews for video generation and live camera, and Perplexity is noted for its web search capability.
Paid vs. Free Subscription
Generally, a paid subscription to these services will get you a better result and come with more safeguards. For example, OpenAI states that it does not train on data submitted through paid subscriptions to ChatGPT Business and Chat GPT Enterprise.
If you use any type of device — cell phone, computer, tablet — then chances are very good that you are using at least one of these agents, paid or unpaid. If so, then you need to practice safe AI.
How to Use AI More Safely
Choose Your Permission
When giving instructions, start with the narrowest permission available. Use non-executing modes like read-only or plan. When you are ready to allow the system to take affirmative acts such as writing, deleting, scheduling, or acting, keep the resulting work in segregated folders.
Don’t Share Sensitive Information
Preserve your native files and pre-run versions. Don’t place sensitive information into your projects unless you have read and understood the product’s disclaimers and warnings, and you are ready to accept the risks.
Keep a List of Sources
If you are using an AI agent to do something important, then keep a list of sources the system accessed, the instructions that you gave, and whether any files were modified or excluded. You never know when you will have to defend the agent’s output.
Be Skeptical
Be skeptical of any result that an AI agent generates. Verify and double-check.
If Generative AI seems confusing, that is because it is. Remember that Socrates once said, “The only true wisdom is in knowing you know nothing.” AI is making all of us wise.
Have Questions About AI and the Law?
AI tools can be useful, but their answers may be incomplete or wrong, especially when you’re dealing with an important legal question. If you need guidance on estate planning, probate, guardianship, or litigation, schedule a consultation with the attorneys at Hammerle Morris.
Virginia Hammerle is an accredited estate planner whose practice includes estate planning, probate, guardianship, and litigation.
This article is for informational purposes only and does not constitute legal advice.
