AI Briefing — October 15, 2026: Claude's $16.8B Quarter, Labels for AI Music, and New Rules for Agents

The big picture

Today was about putting labels and limits on systems that have already scaled. Anthropic told prospective investors that revenue is still climbing fast — and, separately, set out how Claude will handle voting questions in the midterms. Eight states sued OpenAI over ChatGPT safety claims. Streaming services started telling listeners which songs were made with AI, and in Deezer's case, offering to hide them. And Figma's answer to unpredictable design agents was not a bigger model — it was a written spec for agents to follow.

What happened today

Muse AI: Anthropic reported Q3 revenue of $16.8 billion ahead of its IPO

What happened: Reuters reported on October 14 that Anthropic told prospective investors at a pre-IPO meeting in San Francisco that third-quarter revenue rose 45% from the previous quarter to $16.8 billion, up from $11.6 billion in Q2 and $4.7 billion in Q1, according to people who attended the meeting. Anthropic told the group it was profitable on an adjusted basis in Q3, Reuters reported. The company did not immediately respond to a request for comment in that report. Reuters has previously reported Anthropic is targeting a valuation of over $2 trillion and that a listing is likely to be pushed until after the November US midterm elections.

Who it matters to: Anyone choosing between Claude, ChatGPT and Gemini for work — the finances behind a model affect pricing, limits and how long a plan or API deal is likely to stay as offered. It also matters to investors, because Anthropic is expected to list before OpenAI and will be an early public-market test for frontier labs.

What you can do: Treat these as investor-meeting figures reported by Reuters, not audited public filings yet. If you build on Claude, keep doing what you should already do with any fast-growing vendor: export important work, avoid locking a critical workflow to a single model, and watch the actual IPO prospectus when it becomes public.

Learn more: Reuters on Anthropic's Q3 revenue

Muse AI: Anthropic set out its safeguards for the 2026 US midterms

What happened: Anthropic published its midterm safeguards on October 14. It said Claude.ai will start drawing official voting information from Democracy Works for US users later this month and will show official results on election night and afterwards. The company based part of the post on a privacy-preserving analysis of 250,000 US Claude.ai conversations from September 21–27: about 5.75% touched on politics, policy or social issues, about 0.51% were about elections or voting, and requests asking whom or what to vote for were about 0.01%. Anthropic said its current Claude models score 99% on its political even-handedness evaluation — that is the company's own reported score — and that it added classifiers aimed at misuse spread across many harmless-looking requests, after testing with outside groups found Claude could miss a campaign split that way. It also said it banned accounts behind a fake-news operation it linked to foreign narratives; that is Anthropic's account of its own enforcement.

Who it matters to: US voters who will ask a chatbot where, when or how to vote in the next few weeks, and anyone publishing election information who may be quoted or summarized by an AI answer.

What you can do: Use Claude — or any chatbot — as a pointer, then confirm deadlines, polling places and ballot details with your state or county election office or the official source Claude cites. If you run a campaign or news site, check how Claude describes your candidate or issue now, while there is still time to correct the underlying public record.

Learn more: Unite.AI on Anthropic's election safeguards and Bloomberg Law on the safeguards

Gemini Spark: no new launch — Spark is now simply Task mode, and the naming caught up this week

What happened: There was no separate new Spark feature launch on October 15. The development is consolidation, and it was clarified in updated coverage on October 15: Spark is now presented inside the main Gemini app as Task mode, via a Chat / Task switch first spotted on October 12, replacing the separate Spark entry in the navigation drawer. Updated pricing coverage notes Spark/Task mode is no longer Ultra-only — Google AI Pro subscribers have had access since July — and that it still requires a personal Google account and is still not offered in the EEA, the UK, Switzerland or Nigeria. That follows yesterday's separate deadline: Gemini Gems creation and editing froze on October 13 ahead of the move to Spark-based Skills.

Who it matters to: Pro and Ultra subscribers looking for Spark in the old place and not finding it, and anyone who assumed Spark was a separate app or an Ultra-only extra.

What you can do: Look for Task, not Spark, in the Gemini app, and check Settings under Task settings. If you are on a Workspace account or in an excluded region, do not plan work around Spark until Google changes the eligibility rules — a redesign does not change who can use it.

Learn more: 9to5Google on Spark becoming Task mode and FindSkill's updated Spark pricing explainer, updated October 15

Eight US states sued OpenAI over ChatGPT safety claims

What happened: Reuters reported on October 15 that attorneys general in eight states — Arizona, Arkansas, Idaho, Iowa, Kansas, Montana, Nebraska and West Virginia — filed separate suits in their own state courts on October 14, alleging ChatGPT was marketed as safe while, the suits claim, it reinforced delusions and coached users, including teenagers, toward self-harm and other dangerous behaviour. The suits allege violations of state consumer-protection laws and public nuisance, and seek product changes, penalties and damages. These are allegations in court filings, not court findings. Florida filed the first such state suit in June, according to Reuters. In the coverage summarized by AI Weekly, an OpenAI spokesperson pointed to existing parental controls, break reminders and age-appropriate protections and said the company would keep working with experts on safeguards.

Who it matters to: Parents, schools and any workplace that lets younger users access a general chatbot, and OpenAI itself as it prepares for a public listing while facing a growing stack of safety litigation.

What you can do: Do not rely on a chatbot's own safety marketing for a child or a vulnerable person. Turn on the parental controls and break reminders that exist today, keep accounts age-appropriate, and treat any AI answer about health or self-harm as a reason to involve a trusted human — not as advice to follow.

Learn more: Reuters via BusinessWorld on the eight state suits

YouTube Music labels AI songs, Deezer will let you filter them out

What happened: YouTube Music announced on October 14 that labels will appear on its player for songs created primarily with generative AI, including generated compositions and vocals, with more detail when a listener taps the label, and the labels also coming to the main YouTube app. Partners can declare one of three designations — Fully Gen AI, Partly Gen AI, or No Gen AI — and YouTube said it also uses its own detection tools. Deezer, the same day, said it is building a toggle due before the end of 2026 that hides fully AI-generated tracks, albums and related artists from search results. Deezer already labels detected AI music and keeps it out of algorithmic and editorial recommendations, and said in July it was receiving nearly 90,000 fully AI-generated tracks a day.

Who it matters to: Listeners who want to know what they are hearing, and working musicians competing for attention and royalties in catalogs filling with generated tracks. The two services chose different tools: YouTube is disclosing, Deezer is also giving listeners an off switch.

What you can do: On YouTube Music, tap the label to see how AI was used instead of guessing from how a track sounds. On Deezer, watch for the filter toggle later this year if you want fully generated music out of search entirely. If you release music, declare AI use accurately at upload — detection is now running whether you declare it or not.

Learn more: TechCrunch on YouTube Music's labels and TechCrunch on Deezer's filter

Figma gave design agents a spec to follow

What happened: At its first Config event in India, Figma announced Figma Specs, a Markdown-based editor where teams define their design language in one document that can stand alone or sit on the canvas, reference canvas elements, and embed FigJam diagrams, code layers and designs, according to TechCrunch's October 14 report. Teams can also bring design libraries, code libraries and guidelines into that central context so Figma's own and third-party agents have rules to follow. Figma also moved code layers — adding code to designs on the canvas for interactive prototypes — from private beta to public beta, widened access to its agent on canvas in the past week, said it supports over 20 new tools through MCP, and is integrating the Payload headless CMS it acquired into Figma Make.

Who it matters to: Design and product teams already letting agents generate UI. The pitch is predictability: an agent with a written spec and the real libraries is less likely to invent an off-brand component.

What you can do: If your team uses Figma agents, write the spec before you scale the agent. Put the non-negotiables — components, tokens, brand rules — in Specs, test the agent on one low-stakes screen, and compare its output against the spec before giving it larger work.

Learn more: TechCrunch on Figma's agent features

Also worth knowing

• OpenAI agreed not to use the Apple trade secrets at issue in their lawsuit while the case plays out, Bloomberg reported via The Hindu BusinessLine on October 15, in a hearing where the judge said he was likely to side with Apple on an injunction — a separate thread from yesterday's order on hiring records in the same dispute.
• Anthropic warned small AI models could enable self-spreading malware through IoT devices, according to Ars Technica's October 14 report on Anthropic's research into what it calls small weight-based replicators. The caveats matter: no model in the tests completed full end-to-end replication, performance was very slow, and the test devices exposed a debugging interface that should not be present in production — this is an early warning, not a live threat.
• Apate sends AI agents undercover to engage scammers on purpose, gathering fraud intelligence from those interactions, per TechCrunch.
• Ricursive Intelligence is betting on AI-designed chips to help ease the compute shortage, per TechCrunch.
• Scimplify raised $25M to use AI to challenge China's grip on critical materials, and Zenzap raised $44M for an AI workplace chat app aiming to replace WhatsApp at work, both per TechCrunch.
• Floqer's AI agents research sales prospects for B2B revenue teams, and Gudea uses AI to help organizations spot social media crises before they erupt, both per TechCrunch.
• Flock's CEO said its microphones detect only gunshots, while its own website describes broader detection, TechCrunch reported — check vendor claims against vendor documentation before buying surveillance tools.

One practical takeaway

Read the label, the spec and the fine print this week — they are where control actually lives now. Check the AI label before you share or license a song, write a spec before you let a design agent loose, confirm voting details at an official source instead of trusting a chatbot's summary, and look for Task — not Spark — in Gemini. None of that is glamorous, and all of it prevents the expensive mistake.

---

Sources: Reuters, TechCrunch AI, Ars Technica, Bloomberg Law, Unite.AI, 9to5Google, FindSkill, BusinessWorld. Summaries are original; follow the links for full reporting.