Six AI Companies Made a Pledge at the White House. Who Pays When Your AI Tool Gets It Wrong?
On Wednesday, Anthropic, Google, Meta, Nvidia, OpenAI, and xAI all signed something at the White House called the Joint Commitment on Frontier Responsibilities. It's one page and it promises four things. Better internal monitoring of what the models can do. Safeguards so they don't hack or get into systems in ways nobody planned. Outside groups checking the work. A board level committee reviewing progress. President Trump called it "morally binding."
Here's the part I'd read twice. No penalties were announced, and no way to enforce any of it. It's a promise made in a room by the companies selling you the tools. Congress still hasn't passed anything on AI, and the states keep writing their own rules. Lina Khan, who used to run the FTC, said leaning on self-regulation is a clear mistake and that laws we already have, like product liability and consumer protection, should do the job.
I'm not saying these companies are lying. I figure most of them mean it. But a pledge doesn't move any risk off your desk. If an AI tool sends the wrong thing to a customer or leaks your client list, that customer isn't calling Washington. They're calling you.
What Your Vendor's Promise Doesn't Cover
In my workshops I ask who's responsible when the AI gets something wrong. Almost every owner says the vendor. Then we pull up the terms of service together and it says the opposite, usually in a paragraph near the bottom. You own the output, and you're responsible for how you use it. Some of them say the tool can be wrong. A few say it right at the top.
Nobody reads that part when they sign up with a credit card at 10pm. I didn't either, the first few times.
The One Page I'd Write This Week
Make a list of every AI tool somebody at your shop uses, including the ones people signed up for on their own. Next to each one, write a name. A real person who's responsible for checking what it produces before a customer sees it. Not "the team." One name.
Then open each vendor's terms and find the paragraph about who's liable and whether they promise anything on security. Copy it onto your page. It takes about two minutes per tool, and you'll probably be surprised by one or two of them.
Last, send each vendor a one line email: "Do you have a written security commitment, and where do I find it?" Some will answer fast and some will send you a link to a marketing page, and both answers tell you something. File whatever you get next to the list.
The pledge is a good sign and I'm glad it happened. When something does go sideways at your shop, though, you'll reach for your own list with a name on it, because that's the thing that exists on the day you need it.
I run AI workshops and one-on-one AI consultations for businesses around Toledo, Northwest Ohio, and Southeast Michigan, and sorting out who owns what with each AI tool is a regular part of that. If you'd like help putting that page together, send me a note.
Email Jayson