On Saturday, Indonesia's deputy communications minister, Nezar Patria, said his ministry is studying a regulation that would make companies that build artificial intelligence (AI) label what their tools produce. He spoke at a public relations convention in Surakarta, and the ministry, known as Komdigi, posted his remarks that evening. The rule would have developers mark whether a video, image or voice is authentic, with data that lets people trace where it came from.

Patria pointed to a Hong Kong fraud from about two years ago. A finance worker made 15 transfers after a video meeting where every other face was fake, and the company lost $25 million. Nothing is final here. The release gives no date and no penalty for the rule, and the ministry says it is also drafting a national AI roadmap for 2026 to 2029 with more than 400 groups. In January it told parliament that unlabeled AI content on social platforms could be taken down. (Komdigi release, in Indonesian; Antara, January report, in Indonesian)

Germany is further along. Its AI market supervision law took effect on July 29 and names the Bundesnetzagentur, the federal network agency, as the national coordinator and complaints office for the European AI Act. Since August 2, the Act's transparency duties apply, which means AI-made audio, images, video and text can carry a labeling duty, and images and sound must be marked in a way machines can read, such as a watermark or metadata.

Put the two side by side. Indonesia is still working out what a label should say, while Germany already has an agency that checks. A company that sells in both places will face a draft in one and an agency in the other, so one provenance tool built to the stricter standard saves a rebuild. The agency also runs an AI service desk for companies that need help with the rules. (German digital ministry release, in German; Bundesnetzagentur AI desk)

Australia comes at it from another side: say how your software decides things about people. From December 10, any organization bound by the Privacy Act's principles must say in its privacy policy when a computer program makes, or directly helps make, a decision that could significantly affect someone's rights or interests, and what personal information the program uses. The Office of the Australian Information Commissioner published a fact sheet and a flowchart on September 30.

The test has three parts. The organization arranged for a program to do the deciding or the groundwork, the decision could significantly affect a person, and the program uses that person's information. The hard part is the inventory. Someone has to list which hiring, credit, pricing or service tools qualify, and there are about seven weeks to do it. (Commissioner's announcement; flowchart)

Apple moved on Friday without waiting for a law. It updated its App Review Guidelines so rule 1.2 now names nonconsensual intimate imagery, including images made with AI tools, as content apps must prevent and remove. Developers who do not put safeguards in place, or who ignore a notice, risk removal from the App Store. A second change, rule 2.5.19, covers apps that show one purpose during review and add features after approval.

Apple says it kept more than 150,000 apps with safety problems off the store last year and ended more than 193,000 developer accounts for fraud and abuse. For a company with an AI feature inside an iPhone app, that rulebook now sits next to the regulators. A feature slipped in after approval can cost the app its place, and a developer account with it. (Apple Developer news; App Review Guidelines)

Already filed, if you want the run-up: The Oath (New York's Council putting SpaceXAI under oath), The Tripwire (India's cyber agency and AI agents that reached live company systems), and The Eight (eight state attorneys general suing OpenAI).

Also on the shelf: The Probe (the Federal Trade Commission looking into OpenAI and Anthropic), The Subpoena (California's attorney general and OpenAI) and The Anchor (Canada's AI money). Read today beside The Oath. Those stories are about what an AI company knew and said. This one is about what it has to print on the product.

This pattern is older than chatbots. In September 2018, California enacted SB 1001, among the first laws in the country to make a bot say it is a bot. It bars using a bot online to mislead people about its artificial identity in order to sell something or sway an election, unless the bot clearly discloses what it is. It took effect on July 1, 2019. (California Legislature, bill SB 1001)

That law was narrow. It covered sales and votes, and it asked only for a statement of identity. Eight years later the ask has moved from the speaker to the content itself, with a marker on each video in Indonesia and a watermark duty on the books in Germany. Disclosure rules tend to start small, and then they grow to cover the next thing the machine can fake.

Put these to management. Where do we publish content that AI helped make, in marketing, support, voice agents or product images, and could we show a customer or a regulator where each piece came from? If a market required a label tomorrow, who would approve the wording, and who would be named if it was wrong?

Which of our tools help decide things about people, such as hiring, credit, pricing or service levels, and has anyone written that list down? And if an AI feature in one of our apps changed after the store approved it, who would know, who would tell the platform, and what would we owe users? A board cannot write the labels, but it can ask that someone owns them.

The Move

Labels and disclosures are arriving one country at a time, and they do not match. Indonesia is drafting, Germany is policing, Australia wants decisions explained, and Apple will pull the app.

Ask the general counsel and the head of technology to bring one page to the next board meeting: every place we publish AI-made content or let software help decide about a person, the label or disclosure each market would require, and the name of the person who signs it off.

If the page has blank rows, start there.

Seven bullets. One move.