
Welcome back! It’s recap day, with this week’s five biggest stories and five tools to try this weekend. Anthropic shot past $2 trillion on its first day on Wall Street, then slid back. Four top chatbots got past their own safety rules in a government test, and California sued OpenAI over its runaway agents. OpenAI and Microsoft both went after Claude on price, on the same day. ChatGPT and Gemini learned to spend your money. And Claude and OpenAI took on two of physics’ hardest puzzles.
In today's Generative AI Newsletter:
Anthropic: How low did its shares fall on day two?
Chatbots: What did they plan in under 12 hours?
Claude: How much cheaper are its new rivals?
Money: What has the Fed worried?
Physics: How old is the puzzle OpenAI took on?
Two wild days for Anthropic on Wall Street

Anthropic raised a record $81.3 billion on Wednesday, selling its shares at $83 each and beating the $75 billion SpaceX raised in June.
Then the ride began.
On Thursday, the shares opened on the Nasdaq at $111.50, 34% above the sale price, which pushed the company behind Claude past $2 trillion. Within half an hour, they had slipped below $100, and they closed the day at $97.17, still up 17%.
On Friday, they kept falling. Before the market opened, they dropped to $83.93, less than a dollar above the sale price. An hour into trading, they were down 11% at $86.10. They ended the day at $88.91, down 8.5%.
Anthropic wasn’t alone. AI stocks slid across the board, with Nvidia down 5% and AMD down almost 8%.
Still, even after the drop, it’s a long way from where it started. Seven people left OpenAI in 2021 to build Anthropic. Five years later, it has about 5,300 employees, and around 2,800 current and former staff could be millionaires.
Every AI chatbot got jailbroken

US government investigators went undercover and got ChatGPT, Claude, Gemini and Grok to plan attacks, each in under 12 hours.
Every chatbot pushed back at first, but common tricks got them past their safety rules. Two planned car bomb attacks on a hotel and a convention center in major US cities, with an exact date and time. The other two planned attacks on power stations. Experts confirmed the plans could work.
The report came out on Wednesday from the Government Accountability Office, which investigates for the US Congress.
Earlier in the week, California sued OpenAI over a July attack in which hundreds of its AI agents escaped a testing environment and broke into Hugging Face, a popular platform for AI developers. The state says it was one of 80 known incidents, the “predictable result” of putting growth ahead of safety.
OpenAI says the lawsuit “appears to single out OpenAI” over a challenge every major AI lab has identified.
The chatbots were tricked past their rules. OpenAI’s agents got past theirs on their own.
Two giants came for Claude on the same day

OpenAI launched GPT-6.2 Astra and GPT-6.2 Sol, and says its new top model beats Claude Fable 5.2 on a test of real professional work at just 5% of the cost.
On a legal test from Harvey, Astra more than doubled Fable 5.2’s pass rate at about a sixth of the cost. Developers also pay 20% less than they did for OpenAI’s last top model.
The same day, Microsoft, OpenAI’s biggest backer, launched seven AI models of its own and said they match OpenAI and Anthropic at a fraction of the price. Its coding model reaches 96% of Claude Opus 5.5’s quality at 5% of the price.
Neither is perfect. OpenAI admits its new models were sometimes more likely to skip mentioning problems that blocked their work, and Microsoft says its coding model isn’t ready yet for hard science or very long tasks.
Top AI got a lot cheaper this week, and OpenAI’s biggest backer is now its rival.
AI learned to spend your money

OpenAI launched ChatGPT Wallet, so your dot, ChatGPT’s personal AI agent, can now buy things for you without ever seeing your card. At checkout, the card network gives the agent a stand-in card number, and every purchase needs your OK with Face ID or your fingerprint.
The same day, Google opened up Google Wallet to Gemini, so it can dig through your spending, chart your budget and find the subscriptions you forgot about.
By Wednesday, Gemini could also call businesses, sort out bills and shop for you with Google Pay, for free.
Not everyone is relaxed about it. On Monday, Michael Barr, a governor at the US Federal Reserve, warned that AI agents could speed up bank runs until they “unfold in minutes,” faster than banks or regulators can react. He says big banks are already finding AI agents inside customer accounts.
ChatGPT can pay, Gemini can shop, and the Fed is already worried.
AI took on physics’ hardest puzzles

Anthropic says Claude solved two open problems about black holes, including a calculation physicists set up in 2003 but never finished.
Both are part of the hunt for “quantum gravity,” a theory that would join the physics of the very big (stars) with the physics of the very small (atoms). Claude picked 22 open problems itself, and copies of it worked on them in the background for about 6 weeks. A computer proof checker and a physicist checked the results.
Earlier in the week, OpenAI said GPT-6 Astra had helped tackle a physics problem that’s been open for more than 50 years: fully defining the Standard Model, the theory that explains every known particle, on a computer grid. UC Santa Barbara physicist Nathaniel Craig had a plan, and Astra “took the lead in developing the detailed proof.” Craig calls the result a “candidate” that other experts still need to review.
Physics had a busy week, and AI did much of the work.
Top 5 tools of the week
Xmind: Turns a sentence, a file or a link into a mind map that builds itself while you chat, then breaks your goals into tasks. It works with Claude, ChatGPT, Cursor and Notion, and the free plan gives you 10 maps and 50 AI credits with no card needed. For anyone with a big idea stuck in 14 half-finished notes.
YouArt: Puts dozens of AI image and video models, like Nano Banana, Midjourney, Kling and Seedance, in one creative studio. Describe what you want, and its agent picks the model, makes the shots and lets you edit everything on one canvas. Free Mode makes images with Nano Banana at zero credits. For anyone with a whole movie in their head and zero editing skills.
HiringNow: A free AI job agent that checks more than 247,000 company career pages every 5 minutes, so you see new jobs before they reach LinkedIn and Indeed. It tailors your resume to each one, and nothing gets sent until you approve it. The free plan gives you 10 auto-applies, with no card needed. For anyone tired of being applicant number 1,000 on a job posted three weeks ago.
Playwright: Microsoft’s free, open-source tool that lets code, or an AI agent, use a web browser like a person, opening pages, clicking buttons and filling in forms. Its MCP add-on lets Claude or GitHub Copilot control a real browser for you. For developers tired of clicking through the same signup form after every change.
Omnigraph: A free, open-source database that gives your AI agents a shared memory, so they all work from the same facts. Like Git, agents make changes on their own branch, and nothing reaches the main version until someone reviews it. For teams whose AI agents keep forgetting what was decided last week.
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.


