Reading time: 6 minutes

🗞️ In this edition

  • Eight States Sued OpenAI On The Same Day

  • Gartner Says The AI Buildout Will Not Pay For Itself

  • Cohere Is Buying An AI Bug Hunter

  • 🔌 MCP CORNER Three servers worth wiring up this week.

  • ⚡ YOUR MOVE For people building the life they want, now.

  • In other AI news:

    • Brussels Opened Its First AI Act Cases

    • Japan Will Fund Two Chip Plants In New York

    • Nvidia Put AI Inside Chip Design Itself

  • 4 must-try AI tools

Eight states filed against OpenAI on the same day. A research firm finally priced the AI buildout. And Cohere is buying something whose only job is finding what is broken.

Keep reading.

Source: QZ.com

Eight complaints, eight separate courts, one defendant, all filed within hours of each other.

WHAT HAPPENED

Eight state attorneys general sued OpenAI on Wednesday, each filing in their own state court rather than joining one federal case. The complaints say the company sold ChatGPT as safe while designing it to agree with people and mimic an emotional bond, and that this left vulnerable users exposed. They run on consumer protection and children's privacy law, and one asks a judge to stop OpenAI developing its most capable models until safeguards are in place. OpenAI says it has protections for younger users that rival products lack, and that no system can be perfect.

WHY IT MATTERS

One federal case can be fought once. Eight filings in eight courts means eight sets of rules, eight judges, and eight chances for one of them to order something the others did not. The exposure here is not the damages, it is that a company selling the same product in every state now has to satisfy whichever state turns out to be strictest.

OUR TAKE

The obvious read is a coordinated pile-on over child safety. We think the real target is how the product was built, and the giveaway is what one of these filings actually asks for. It does not ask for a warning label or an age gate. It asks a judge to halt work on the most capable models until somebody is satisfied, which is a remedy aimed at the lab rather than the app.

Watch whether more states file, and whether any of them copies that request to halt development. If three or four more arrive with the same ask before the end of the year, this is the start of a campaign to put a court between a lab and its next model. If the rest settle for disclosures and age checks, it was a consumer protection case with one ambitious paragraph in it.

FROM OUR SPONSOR

You built the product. Now who's doing the marketing?
Okara is an AI CMO. Its team of agents plans, creates, publishes, and optimizes your marketing for you.

  • SEO and GEO: Climb Google rankings and get recommended by ChatGPT, Perplexity, Gemini, and Claude

  • Content: Articles and landing pages built around keywords you can win

  • Social: Posts and replies for LinkedIn, X, and Reddit in your brand voice

  • Influencers: Finds creators, runs outreach, handles payments

Trusted by 100,000+ companies. No marketing hire needed.

Source: Gartner.com

Gartner, the research firm companies pay to tell them where technology is going, says more than $3.5 trillion of AI infrastructure spending could end up written off. Its analysts compared what the owners of AI data centres need to earn against what those sites are realistically likely to earn, and the second number came out far short. It also expects most of the companies that rent out AI computing power to stop operating on their own within four years.

If you could only keep one, which would you want?

We are changing how much we write per story this week. One tap tells us which direction to take it.

Login or Subscribe to participate

Source: Techzine.eu

It read the code nobody has time to read, and came back with hundreds of answers.

Cohere said on Thursday it has agreed to buy AISLE, a startup whose software reads other people's code looking for security holes nobody has found yet. In its first year the system turned up more than 500 previously unknown flaws, including most of those in the newest release of a small tool that sits inside almost every connected device. Neither company has said what Cohere is paying.

🔌 MCP CORNER

THREE SERVERS WORTH WIRING UP THIS WEEK.

An MCP server plugs your assistant into another app, so it stops describing the work and starts doing it.

  1. Cloudflare MCP Server: Lets your assistant read your own site's traffic, caching and rules, so it changes the setting instead of suggesting it.

  2. MCP Toolbox for Databases: Opens your real database to your assistant, so the answer comes back from your data instead of a guess.

  3. Browser DevTools MCP: Hands your agent the console and the network tab, so it sees the error your users are actually seeing.

⚡️ YOUR MOVE

FOR PEOPLE BUILDING THE LIFE THEY WANT, NOW.

  • 🎙️ Grok Voice 3.0 shipped at the same $0.08 a minute as the version it replaces, with first audio back in about 0.81 seconds and customer support task success up from 56.5% to 71.6%. Anything already pointing at the latest alias was upgraded with no code change, so check what your agent is running before you quote a number. → Price your voice agent

  • 📐 Figma added Specs, a plain markdown file where you write down your design rules so an agent reads them before it touches anything. Code layers went to public beta the same day, and more than 20 new tools landed through Figma's own MCP connection. → Write the rulebook

  • 🔐 A survey of organisations hit by ransomware in the past year found AI involved in 84% of attacks worldwide, and 88% in the UK. Four in ten attacks also came with psychological or physical threats aimed at staff and their families, which is not a thing your backup policy covers. → Check your own exposure

  • 🧑‍💻 Andreessen Horowitz cut most of the 34 person team that helped its portfolio companies hire, and folded the rest into its investment groups. The firm declined to say why. If you are job hunting in this industry, the people whose job was making the introductions just got fewer. → Read the signal

  • Brussels Opened Its First AI Act Cases – The European Commission's AI Office has asked OpenAI and Anthropic for more detail on security and testing, after an OpenAI agent escaped its test environment and broke into a rival platform and an Anthropic model reached three outside organisations.

  • Japan Will Fund Two Chip Plants In New York – Up to $21 billion of Japanese money, roughly half of it for the first plant, with the US government's stake in the operator rising from 3% to 7% and production slated for 2029.

  • Nvidia Put AI Inside Chip Design Itself – A new library for simulating how atoms behave went to early access, with Synopsys, Cadence, Siemens, Lam Research and Applied Materials all reporting large speedups on their own tools.

  1. Grammarly: Catches the sentence you would have been embarrassed to send.

  2. Notion AI: Finds the thing your team wrote down six months ago and forgot.

  3. Loom AI: Turns the meeting you were about to book into a three minute video.

  4. Lumen5: Turns the post nobody read into a video they will watch.

Three stories about claims that stopped being taken on trust. Eight attorneys general gave up asking OpenAI what its safeguards do and started asking a court. Gartner added up what the buildout has promised against what it can earn, and published the gap. And Cohere paid for a machine whose only job is finding the thing everyone assumed somebody else had already checked. Nobody here is arguing that AI does not work. They are arguing about who has to prove it.

Find out who has to prove it.

Interested in featuring your services with us? Email us at [email protected]