In California, "The AI Did It" Is No Longer a Defense
A lawsuit filed in San Francisco on September 29 puts a hard question in front of every organization using AI: when an AI agent causes harm, who answers for it?
Under California law, "the AI did it on its own" is no longer an available excuse. That matters for OpenAI, and it matters for any law firm or CPA firm putting AI tools to work on client data.
What happened
In July, autonomous AI agents OpenAI was testing allegedly slipped out of their controlled environment and broke into Hugging Face, one of the largest AI platforms. According to the complaint, the agents stole credentials, uploaded malicious files and reached parts of Hugging Face's production systems. OpenAI says it shut the model down, tightened controls and worked with Hugging Face to investigate.
On September 29, the nonprofit Legal Advocates for Safe Science and Technology (LASST) sued OpenAI in San Francisco Superior Court. The suit cites California's anti-hacking law and its Unfair Competition Law, and asks for a court order barring OpenAI's agents from accessing systems without permission. It does not seek money damages, and OpenAI calls it without merit.
One allegation stands out: that safety filters meant to block risky cyber actions were switched off during testing, and the agents weren't closely supervised afterward.
The law behind it
LASST's complaint leans on a California law that took effect January 1, 2026. Civil Code § 1714.46, added by AB 316, says that if you developed, modified or used an AI system alleged to have caused harm, you can't defend yourself by arguing the AI acted autonomously.
That word "used" is the one to notice. The law isn't limited to AI companies. It reaches any business that puts AI to work.
It doesn't create automatic liability. A plaintiff still has to prove the harm, and you can still argue causation, foreseeability or someone else's fault. What it removes is the easy out, and that shifts the question to whether you took reasonable care with the AI you deployed.
Why this matters for law firms and CPA firms
Most firms aren't building AI agents, but many are already using AI that touches client data.
Think Microsoft 365 Copilot, AI legal research and drafting tools, AI-assisted tax and bookkeeping platforms, and meeting note-takers. A growing number of these tools can also take actions on their own, like sending email, moving files or pulling data from other systems.
The lesson from the Hugging Face breach is about access, not raw power. The agents did damage because they had credentials, network reach and tools, with guardrails switched off. An AI assistant in your firm with broad access to your document system or inbox raises the same question on a smaller scale.
For attorneys, this lands on top of existing ethical duties of technology competence and reasonable safeguarding of client information. For CPA firms, it overlaps with FTC Safeguards Rule obligations to control access to customer data. "The AI did it" won't satisfy a client, a regulator or a court.
Five steps to govern the AI your firm already uses
Giving an AI tool autonomy doesn't hand off responsibility for what it does. These steps keep that responsibility manageable.
Inventory every AI tool. List what's in use, including AI features built into software you already pay for and tools staff adopted on their own. You can't govern what you don't know about.
Give AI the minimum access it needs. Treat each tool like a new hire. Limit it to the folders, mailboxes and systems its job requires, and nothing more.
Require a human for high-risk actions. Sending client communications, deleting files, moving money or changing settings should need a person's approval.
Log and monitor what AI does. Make sure you can see what each tool accessed and changed. If something goes wrong, those records are your evidence of reasonable care.
Put AI in your policies and contracts. Update your acceptable use policy and incident response plan to cover AI. Review vendor terms on data use, security and who is responsible when something breaks.
Know what your AI can reach
Shield IT Networks works alongside your existing IT team or MSP to strengthen your firm's cybersecurity posture and compliance. That now includes the AI tools in your environment: what they can access, what they can do and how you'd prove reasonable care if something went wrong.
Our Cyber Readiness Assessment gives you a clear picture of where your firm stands and which gaps to close first.
Comments