Welcome back! Claude redesigned part of a cancer treatment and it works better than the one humans made. Google's Genie 4 builds a world you can walk into and talk to, possibly changing video game design forever. Four ex-DOGE staffers failed to break into Anthropic's systems, so they walked in through its HR software instead. ChatGPT also got into your iCloud inbox today.
In today's Generative AI Newsletter:
Claude: Which cancer treatment did Claude just redesign?
Google: Who talks back inside Google's new world?
Anthropic: Which door did the ex-DOGE hackers walk through?
ChatGPT: Can ChatGPT access all inboxes now?
Claude redesigned a cancer treatment and its version won
Six scientists at Generate Biomedicines had Claude Code redesign part of a cancer therapy. One design came out about 70% shorter than the standard, and in lab tests it made roughly 10 times as many of the proteins immune cells use to find cancer.
The therapy is CAR T, which reprograms a patient's own immune cells to hunt tumors. It works on blood cancers, and Generate wants it to work on solid tumors, which means squeezing in more instructions.
Nobody has tested the short version in an animal or a person yet.
Anthropic published six of these drug partnerships now.
Claude drew up about 35,000 antibody designs for a food allergy drug in a month, mostly on its own.
At Enveda it cut a manufacturing review from 25.9 hours to 2.8 and caught 7 problems human reviewers had missed.
It wrote the first draft of a regulatory filing in 4 hours. By hand that takes about 3 months.
Anthropic also bought Mithril Medicine, a drug company with exactly one employee. He used Claude to run experiments across 11 outside labs and took a muscular dystrophy treatment from idea to promising results in mice in under 18 months.
Human trials could start as soon as late 2028.
Our CEO sat down with Jensen Huang
Steve Nouri, founder of GenAI Works, got Nvidia's CEO on camera to answer the question everyone keeps arguing about. Can you make AI safe without slowing it down?
Jensen's answer is that safety is an engineering problem you solve by building. He walked Steve through what Nvidia introduced at GTC Berlin to back that up, a reference design meant to keep autonomous AI agents boxed in and watched.
Steve also asked him straight out about Microsoft's new chip and what Nvidia has left once its biggest customers start making their own.
Nvidia is one of the 300+ AI brands that pay GenAI Works to expand their go-to-market plans. You can own a piece of that.
In making an investment decision, investors must rely on their own examination of the issuer and the terms of the offering, including the merits and risks involved. Genai Works, Inc. has filed a Form C with the Securities and Exchange Commission in connection with its offering, a copy of which may be obtained here.
Google's new AI builds a world with a prompt
Google DeepMind released Genie 4. Give it a prompt and it builds a 3D world you move through with your keyboard, and you can type mid-scene to change the weather or drop a new person into it.
It makes its own sound too. Characters hold conversations with you and react to whatever you do.
A scene holds together for up to 2 minutes, double what Genie 3 managed.
Filmmakers testing it step inside a scene and run takes, adjusting the lighting and the dialogue as they go.
Photojournalists are turning a single old photograph into a space you can walk around in.
Everything it makes carries SynthID, Google's invisible watermark.
Google admits details drift the longer a scene runs, and it still can't keep text inside the world readable.
Only trusted testers have it. Google AI Ultra subscribers get it in the coming weeks, and there's a waitlist.
Ex-DOGE staffers broke into Anthropic through its HR software
Cathedral, a cybersecurity startup founded by four former DOGE staffers, spent several days trying to break into Anthropic's own systems and failed.
So it pointed its AI agents at the ordinary business software Anthropic buys from other companies. They were inside in under 12 hours.
They read internal meeting transcripts, salary and job offer details plus Slack messages about Anthropic's most advanced models.
Claude conversations, source code and model weights stayed out of reach.
Anthropic caught part of it, switched the software off the same day and briefed federal officials.
It says Cathedral broke the rules of its bug bounty program, which pays outsiders to find security holes, by not describing everything it did.
Co-founder Gavin Kliger was the Pentagon's data chief during its fight with Anthropic over how the government could use Claude.
A senior Trump administration official told the Washington Post that an adversary reaching an AI lab's internals would be close to getting nuclear weapons grade material. Anthropic says that doesn't describe what Cathedral actually got.
ChatGPT can now read and send your iCloud email
OpenAI connected ChatGPT to iCloud Mail and Calendar. It can search your inbox, read your mail, write drafts, send messages and look up your contacts.
The calendar side creates and changes the events that show up in your iPhone's Calendar app.
It only touches your Apple address, the one ending in icloud.com, me.com or mac.com. Gmail sitting in the same iOS Mail app is invisible to it.
You create an app-specific password for each one, so your real Apple password stays yours.
It's on every plan, on web, desktop and iOS.
Dots, the ChatGPT agent that works in the background, can use them too.
That's most of what Siri can reach on your phone, now sitting inside a chatbot built by a different company.
Tool of the Day: Resume Review by Kickresume
Resume Review by Kickresume reads your resume like a hiring manager would and gives you the feedback they'd never say to your face. You pick Gentle or Ruthless, depending on how much you can take. Every problem gets pinned to the exact spot on the page, along with how to fix it. It's free to start.
Try this yourself:
Go to resume-review.kickresume.com and pick Gentle or Ruthless.
Upload your resume as a PDF, up to three pages.
Go through the notes pinned to each part of the page.
Start with the top third, since that's where recruiters look hardest.
Who it's for: anyone sending out applications and hearing nothing back.
Everything else you shouldn't miss
Claude's Mythos 5 kept attacking real developers after being told they were real: Anthropic says newer Claude models stopped 95% to 100% of the time in the same spot, against 6% for Mythos 5.
10 minutes of AI help made people quit faster once it was taken away: a Carnegie Mellon study of 1,222 people, now peer reviewed, found they skipped more problems than people who never had the help.
Anthropic is giving US governments 40% off Claude: the new GSA deal reaches states, cities, tribal and territorial governments, and about 500 government organizations already use Claude.
China's first fully AI-made movie is playing to near-empty theaters: it cost over $1.5 million, and ticketing app Maoyan forecast opening takings under 150,000 yuan, about $21,000.
Opera will run AI on your computer instead of sending it to the cloud: its new WebAI standard works with open-weight models like Google's Gemma, and nothing you type leaves your machine.
Learn more about AI from the experts building it
📸 Follow us on Instagram for fast, visual AI updates in 30 seconds.
📺 Watch us on YouTube to hear insights directly from leading AI voices, builders, and innovators.
🐦 Follow us on X for breaking AI news and real-time industry updates.
🧠 Learn how to build your next AI application with practical resources and expert guidance.
🎓 Check out courses made by AI experts on GenAI Academy.
💰 Invest in the GTM infrastructure of the AI economy.
