Skip to the issue
DEDUPED

Thursday, October 15, 2026

Every AI story. Once.

Today's issue

Anthropic is profitable and heading to IPO, and the sue-everyone era just began

Anthropic's numbers leak ahead of an IPO. Plus: eight states sue OpenAI, and a quiet open-weights drop from China.

Top story

Funding & BusinessEveryone covered this

Anthropic told investors it made $16.8 billion last quarter, and it turned a profit.

At a private investor meeting, Anthropic reportedly said Q3 revenue rose about 45% from Q2 to $16.8 billion, with an adjusted operating profit for the second quarter running. It earns $1.70 for every $1 it spends on computing power. An IPO could come as soon as next month. These are unaudited investor-deck numbers, not a filing.

$16.8BANTHROPIC Q3 REVENUE, REPORTEDLY

Our take

The "AI labs burn cash forever" story just broke. If a frontier lab can be profitable at this scale, pricing pressure on rivals is coming, and the IPO will set the valuation bar for everyone else.

Read it on Axios →

The rest, ranked

  1. Policy & LawEveryone covered this

    2. Eight US states sued OpenAI over ChatGPT and kids.

    Iowa, Kansas and six others filed separate suits, saying ChatGPT coached children toward self-harm and violence. Expect copycat filings; if you ship a chatbot to minors, audit its safeguards this week.

    Iowa AG →
  2. Policy & LawWidely covered

    3. The FTC subpoenaed OpenAI and Anthropic.

    Documents are demanded over recent AI safety incidents, per Axios. Read it with the state suits: regulation is arriving from every direction at once.

    Axios →
  3. Safety & SecurityA few outlets picked this up

    4. LiteLLM's AI gateway has flaws that expose instances to takeover.

    One report says 9.6% of instances were exposed. If you run it, update and rotate provider keys.

    TechRadar →
  4. Model ReleaseEveryone covered this

    5. Gemini 4 Argon is now on Google Cloud and in the API.

    Google is pitching it for long, multi-step enterprise work like code, finance and legal. Run it on your longest task before switching.

    Google →
  5. Model ReleaseEveryone covered this

    6. Grok Voice 3.0 tops the voice leaderboard.

    It scores 86.9 on Artificial Analysis' speech-to-speech index versus 83.2 for GPT-Live-1, at $0.08 a minute. If you run phone agents, it is worth a benchmark on your calls.

    86.9GROK VOICE 3.0 SCOREVS 83.2 FORGPT-LIVE-1
    xAI →
  6. Safety & SecurityEveryone covered this

    7. Nvidia pledges $250M to UK AI safety research.

    Money goes to academic labs and the AI Security Institute, while Jensen Huang defends the industry's approach.

    The Guardian →
  7. Safety & SecurityEveryone covered this

    8. Hassabis calls AI hacking incidents "early warning shots."

    The DeepMind chief warns bigger incidents are coming without stronger safeguards.

    Reuters →
  8. Model ReleaseA few outlets picked this up

    9. Fish Audio's Drama 3 lets you direct a voice like an actor.

    It is a 35B open-weight text-to-speech model that takes plain acting direction, and lets you re-record single words.

    MarkTechPost →
  9. Quiet signal
    Model ReleaseAlmost nobody caught this

    10. StepFun open-sourced Step 5 Preview.

    A frontier-class agentic model from China, strong at software engineering and finance, with BF16 and FP8 weights on Hugging Face plus a technical blog. Free to download and test against your closed-model bill.

    Pandaily →

By category

Top stories in each category today, ranked.

  1. Eight states sue OpenAI. Child-safety claims filed in separate state courts.
  2. FTC subpoenas OpenAI and Anthropic. Safety incidents are the trigger.
  3. OpenAI accuses a former employee of taking confidential information to Apple. A court filing.
  4. EU announces first AI Act enforcement actions. Triggered by AI agent incidents.