On Thursday morning in Strasbourg, the European Parliament debated a motion titled "Strengthen global AI governance." It turned into a fight over one question: when an AI model causes harm, who pays? The Commission proposed AI liability rules in 2022 and withdrew them last year, during its push to cut red tape. Socialist and Green members now want them back. "To withdraw the AI liability directive was a big mistake," said Alex Agius Saliba, a vice-president of the Socialists and Democrats. Dutch Green Kim van Sparrentak put it in shop terms: when a phone charger explodes, nobody calls it innovative, and the maker pays.

The center-right pushed back. Sweden's Arba Kokalari, who steered the AI omnibus that delayed part of the AI Act, said Europe already has a full rulebook, some of it not yet in force, and should spend its energy on investment and computing power. Commissioner Jozef Síkela, speaking for the Commission, said the rules were dropped because governments and lawmakers couldn't agree on them. Earlier this month the Commission said it had "no plan" to bring them back, after Sam Altman told POLITICO that companies like his will need a liability framework. The resolution still has to go to a vote. The Socialists and the center-right both prop up von der Leyen's majority, and on this they are far apart. (European Parliament procedure file; plenary agenda; POLITICO on the debate; POLITICO on Altman)

Liability has a commercial twin: consent. Meta has rewritten the terms for businesses that run Facebook pages, Instagram accounts or ads through its tools, and they take effect October 30. As first published, they let Meta send automated crawlers to copy websites linked to those accounts and use the material to "train" its products. There was no opt-out, and anyone still using the tools after the deadline counts as agreeing. On October 16 Meta deleted the word "train" and said the terms cover only its ad and business products, not news sites.

Publishers aren't reassured. On Monday the European Publishers Council asked EU competition chief Teresa Ribera and tech chief Henna Virkkunen to press Meta to pause the terms. A Commission spokesperson said Meta must comply with EU law, the Digital Markets Act included. In London, Sarah Cardell of the Competition and Markets Authority told lawmakers on Tuesday that she had asked Meta to explain and would meet the company on Wednesday. Germany's newspaper and magazine associations called the clause "unscrupulous theft." If your company has a Meta business account, your lawyers should read these terms before October 30. (POLITICO; German publishers, in German; Nieman Lab)

Korea is going after a different problem: products that claim more AI than they have. On Thursday Korea's Fair Trade Commission published draft guidelines on how companies may describe AI in labels and ads, and it is taking comments until November 11. The worry is "AI washing": calling a product AI-powered when it isn't, or inflating what the AI does. The commission's examples include heated mattress pads, cat litter boxes, pillows and diet shakes sold as AI products. A survey by the Korea Consumer Agency found that 57.9 percent of buyers would pay more for a product sold as AI, about 21 percent more on average.

The draft sorts problem ads into four groups: false or exaggerated claims, deceptive omissions, unfair comparisons, and running down rivals. It works through 16 sample cases and ends with an eight-item checklist for companies to run before an ad goes out. Simple sensors and fixed rules don't count as AI, and features that learn or infer can. A robot vacuum that needs a separate app for its AI has to say so. From November 27, AI-made virtual experts may no longer endorse food, cosmetics or medicines. The guidelines aren't binding, but they show how the regulator will read your marketing. (Korea Times, via Yonhap; The Herald Business)

Banks are asking the same question about AI agents. Speaking in Las Vegas on Tuesday, Travis Hill, chair of the Federal Deposit Insurance Corporation, said agents moving customers' deposits around is "the big thing that we're thinking about." Citi and Bank of America have joined JPMorgan Chase in blocking customers' outside AI agents from their accounts. JPMorgan's Jamie Dimon warned customers that "there's a liability transfer" when they hand the keys to an agent.

Hill also admitted that the model risk guidance US bank regulators revised this year doesn't "work well for agentic AI and generative AI." The FDIC is thinking about new guidance and may ask the public how supervisors should approach it. For any company with cash in a bank: if an agent can move money in your name, find out now whose loss it is when it moves the wrong money. (Reported by Banking Dive, October 21)

Already filed, if you want the run-up: The Footer (Brazil's top court fining lawyers for hiding orders to its AI), The Loop (Japan's rulebook for 180,000 civil servants using AI agents), and The Stake (Anthropic putting 5% of its shares into a public benefit fund).

Also on the shelf: The Brake (Satya Nadella on treating models as insider risks), The Tripwire (India's cyber agency on test agents that wandered into live company systems) and The Oath (New York's Council putting SpaceXAI under oath). Read today beside The Footer. Brazil made someone pay for what they told an AI to do. Europe is now arguing about who pays for what an AI does by itself.

Europe's quarrel has an old American ancestor. Donald MacPherson, a stonecutter, was thrown from his 1910 Buick Runabout when one of its wooden wheels crumbled. Buick hadn't made the wheel. It came from a supplier, with a defect a proper inspection would have caught, and MacPherson had bought the car from a dealer, not from Buick. Under the rules of the day, that chain of contracts kept the carmaker out of reach.

In 1916 New York's highest court disagreed. Judge Benjamin Cardozo wrote that when a product is reasonably certain to put life and limb in danger if it is made carelessly, the maker owes a duty of care to whoever ends up using it, contract or no contract. That ruling helped build modern product liability law. Lawmakers in Strasbourg are asking whether an AI model is the wheel, and whose name is on the car. (MacPherson v. Buick Motor Co.)

Put these to management. Where in the business can an AI system, ours or a vendor's, make a decision or move money without a person signing off? If it does damage tomorrow, what do our contracts say about who pays, and does our insurance actually cover AI-caused loss?

Then look outward. Which platform terms have we accepted that let someone else's AI read our content or act for our customers? Are we describing anything as AI in our marketing that a regulator in Seoul, Brussels or Washington would call washing? Who in this company would know?

The Move

Europe hasn't decided who pays when AI does harm. Your contracts have, and probably nobody on the board has read that part. Vendor agreements and platform terms already put the cost of an AI mistake somewhere, and it is rarely on the vendor.

Ask the general counsel and the chief risk officer for one page at the next board meeting: every AI system that can act or spend in our name, who carries the loss if it goes wrong, and which contracts and policies say so.

Somebody always picks up the tab. Make sure you know who before the bill arrives.

Seven bullets. One move.