As Artificial Intelligence grows in usage and influence it is time for those working in digital forensics, criminal investigations and incident response to be more aware of its effects and capabilities. Steve Whalen wrote a very informative article that I think every investigator, incident responder, and forensic analyst should read. Not just read but explore the ideas presented and begin preparing for the inevitable “The AI Did It!”
Historically speaking this isn’t really new. In the past decades a typical criminal defense was, “I didn’t do it!” Or “I have no idea how that got onto my computer!” Or it is a “fake image” and can not be criminally prosecuted. Law enforcement and other defenders of society eventually managed to catch up and begin solving these problems in court and holding those criminals responsible, but now AI is in the picture.
For example: “We’re not talking about ChatGPT answering questions anymore. The latest generation of AI agents can:
Manipulate your file system: read, write, move, rename, and delete files across your entire directory structure
Control your web browser: navigate websites, fill forms, extract data, interact with web applications
Access sensitive applications: read and send Messages, manage Apple Notes, control Chrome tabs
Execute system commands: run AppleScript, bash scripts, and automate macOS tasks
Make autonomous decisions: chain multiple operations together without constant user oversight”
As you can see, the effects on a suspect’s file system are immense and often unknown and unseen. Mr. Whalen describes this as a “paradigm shift.” These agents do not perform like typical software, they are operating on their “own” predetermined” decisions and can perform multi-step operations, that in the past would require intensive end user interaction.
So, what do we do? How do we prepare for the inevitability of the claim that “AI did it!”
- First, learn what the AI is logging and not logging. Whalen’s article provides an excellent summary of some places to view these logs and the information that can be gleaned from them.
- In some instances, the AI must request permission to perform sensitive operations, such as file deletions or removal. But we must keep in mind that many end-users allow this without knowing exactly what it could entail!
- Search for the session context and user intent. As Harlan Carvey has written, these logs become a “forensic artifact.” A part of the procedure and not the end all, be all, of forensic findings.
- Many AI programs produce timestamped files and log that information which would allow a forensic analyst to compile a timeline of the AIs activity. This could prove very insightful and helpful in an investigation.
The above items are just examples of data and information that analysts need to become increasingly aware of when examining systems. We must learn AI, learn how it logs, learn what it does. But most importantly, we need to keep in mind that “AI” only act with “delegated authority!”
“The user must:
Authenticate and set up the AI platform
Explicitly grant access to specific folders and capabilities
Approve permission requests for sensitive operations
Initiate sessions with stated intent
Maintain the ability to monitor and stop operations in real-time
When someone says “the AI did it,” what they’re really saying is “I authorized an AI to do it, but I don’t want to take responsibility for what I authorized.”
Be ready for AI to be used as an excuse for illicit behavior and activity. Be ready to prove the course of conduct! Read the article and begin learning AI and its forensic artifacts!
Leave a Reply