AI news for Thursday, October 15, 2026: Eight states sue OpenAI, Anthropic's $16.8B quarter

No new model today. The pressure on AI moved to courtrooms and ledgers, and Figma shipped the one thing you can try this week.

Policy Safety3 min
Eight states sue OpenAI over ChatGPT safety for kids
Iowa, Kansas, Nebraska, West Virginia and four more states sued OpenAI on Oct. 14, alleging ChatGPT endangers children and lacks age checks. Eight suits in one day means the fight is now over how the product is built.
Why it matters If your app puts a chatbot in front of anyone under 18, the states just told you what they will look for: age verification and real safeguards around self-harm. Build both before a court decides how they should work.
SRCWest Virginia Attorney General
Business3 min
Anthropic told investors its Q3 revenue hit $16.8B
Anthropic told prospective investors its third-quarter revenue reached $16.8 billion, up from about $11.5 billion, and that it was profitable on an adjusted basis, the FT reports. Growth is slowing. The profit is the bigger news for anyone who pays Anthropic.
Why it matters If you build on Claude, your supplier now says it makes money and will soon answer to public shareholders. Both shape what you pay. When the public S-1 lands, read it like a supplier audit: customer concentration and compute costs tell you more about your next price change than any launch post.
SRCYahoo Finance (relaying the Financial Times)
Agents Products3 min
Figma launches Specs so coding agents read the brief
Figma released Specs in open beta for paid plans on Oct. 15, a doc on the canvas that its own agent and outside coding agents can read through MCP. Agents fail on missing context, and Figma wants to be where that context lives.
Why it matters If you already paste Figma links into a coding agent, Specs gives the decisions that never made it into the frames a place the agent can read. Try it on one feature this week and compare the agent's first pass.
SRCFigma
Hardware Business3 min
TSMC's profit jumps 68.7% and it guides Q4 higher
TSMC's third-quarter net income rose 68.7% to NT$763.06 billion, and it guided fourth-quarter revenue to $52.3 to $54.3 billion. Whatever is rattling AI investors, the foundry is guiding up.
Why it matters TSMC's 3nm and 2nm chips made up 41% of its wafer revenue, and it expects to sell more next quarter at nearly the same margins. For you, that means more advanced silicon coming online, though a 66.4% gross margin says cheaper chips are not part of the deal.
SRCTSMC
Creative Products2 min
YouTube Music now labels songs made mostly by AI
YouTube Music began labeling songs created "primarily using generative AI" on Oct. 14, working with music partners and its own detection tools. A day after Deezer promised a filter, labels look like the floor, and the line YouTube drew leaves the hard cases open.
Why it matters If you make music with AI tools, how much the model did now decides whether listeners see a label. Keep a record of what you wrote and what the model generated, because YouTube is relying on its music partners and its own detectors to make the call.
SRCYouTube

ShareX LinkedIn Hacker News Email

Prompt of the day

Write a feature spec your coding agent can actually follow

› You are a senior product engineer helping me write a feature spec that an AI coding agent will build from.

The feature: [describe the feature in two or three sentences]
The product: [what your product does and who uses it]

Interview me first. Ask up to eight short questions, one at a time, about the decisions an agent cannot guess: the user's goal, edge cases, empty and error states, the data it reads and writes, permissions, what must not change, and how we will know it works. Wait for each answer before the next question.

Then write the spec in Markdown with these sections:
1. Goal (one paragraph)
2. Who uses it and the job they are doing
3. Requirements (numbered, each one testable)
4. Edge cases and error states
5. Data and APIs touched
6. Out of scope
7. Acceptance tests (Given / When / Then)
8. Open questions

Rules: plain language. Every requirement specific enough that two engineers would build the same thing. Mark anything you assumed with [ASSUMPTION] so I can confirm it. End with the three decisions most likely to make an agent build the wrong thing, and the question I should answer for each.

Open in Magai

Works in any AI chat. In Magai, run it against several models at once.

Get every edition

This one arrived in inboxes at 6am ET.

Eight states sue OpenAI over ChatGPT safety for kids

Anthropic told investors its Q3 revenue hit $16.8B

Figma launches Specs so coding agents read the brief

TSMC's profit jumps 68.7% and it guides Q4 higher

YouTube Music now labels songs made mostly by AI