AI Legal Signal Brief

Directors are not held liable for having an imperfect AI oversight system. They are held liable for not building one and knowing it. The bar is bad faith — and the least expensive way to stay well clear of it is a documented system showing the board actually looked.

Write the post that corrects the reflex reading. Most coverage will land as boards being off the hook. Open there, then turn it: the reason this went the board's way is precisely why an AI oversight record is worth building now, while it is still cheap and voluntary rather than reconstructed under discovery.

Substitute model deployment decision for the operational decision in that article and it is about nearly every company deploying AI right now. AI is being governed as an engineering fact instead of an enterprise risk — which means the decision carrying the most value is the one with the thinnest record a board could ever point to.

Turn the five properties into a working session, not a slide. Give each executive one live AI use case and ten minutes to draft the one-page pre-deployment risk position, then have the room grade each other's drafts against a single test — is it specific enough to be wrong? Most first drafts are not, and discovering that in a workshop is considerably cheaper than discovering it in a deposition.

Eighteen AI-related securities suits have been filed in 2026, and the newest one has nothing to do with the AI. Investor enthusiasm for AI does not invent a new category of disclosure risk so much as it amplifies the oldest one — a growth story resting on customers nobody diligenced.

Post: The AI lawsuit that isn't about AI. Draw the distinction executives routinely blur — AI washing means overstating what the model does; AI-adjacent means ordinary fraud amplified by AI enthusiasm. Being able to tell the two apart in a board meeting is a credibility marker. Podcast version: the AI premium cuts both ways, because the same narrative that lifts the multiple is what attracts the short seller.

This is the data-provenance question with a stock chart attached. The allegation is not that someone breached them — it is that their own data supply chain was the problem, and when a third party switched it off, half the market value went with it.

Post or talk segment: three questions to ask any AI vendor about where the data came from — whose consent, what chain of custody, and what happens to your service if their supply is cut off. The third question is the one almost nobody asks and the one this case answers. Natural bridge into vendor indemnity: an indemnity clause is cold comfort when the vendor's entire pipeline goes dark in a week.

The data-broker layer is where much AI training and enrichment data quietly originates, and California is now the most active enforcer sitting on top of it. If your model got measurably smarter last quarter, someone in your company should be able to say where that data came from.

Short explainer: your AI vendor's vendor. Then the workshop version — put one live use case on the board and have the room draw the data chain one layer deeper than the contract they actually signed. Most rooms stop at layer one, and that gap is the entire exercise.