Issue no. 3

 

Bleecker Street Digital, on a postage stamp: Agentic Resources / Private AI Infrastructure. Your personal agents. Your own data. Sleep peacefully tonight.

Friday, November 20, 2026

 

The Bleecker Street Daily

A bundle of news to keep you informed of all things open-source and AI.

 

Anthropic Launches Historic IPO

Anthropic completed the largest IPO of all time and shares began trading yesterday, November 19th. Anthropic said it sold 980 million shares, including some from its founders. The company also planned to raise $26 billion through a private placement, bringing its total haul from the offering to $107 billion. Anthropic had set a preliminary price range of $70 to $78 a share but raised it amid high demand from investors. Anthropic’s shares are trading under the symbol ANT. 

Anthropic’s offering is the first time that public market investors will have the chance to invest in one of the A.I. industry’s “frontier labs,” which are developing the core models underlying the boom. The company’s offering prospectus showed a fast-growing business alongside a narrative of future growth and abundance, including curing diseases and expanding economic prosperity. Revenue soared to $33 billion in the first nine months of the year from $41 million in 2023. 

The company has had an astonishing rise in just five years. Founded in 2021 by a group of former OpenAI employees, Anthropic and its Claude chatbot were a somewhat obscure second place player to OpenAI’s ChatGPT. But Anthropic’s focus on business tools, including Claude Code, which automates software development, became increasingly popular over the last year. Source

Models & Research

An Anthropic mathematician claims he has proved the Riemann hypothesis — working with Claude. Levent Alpöge, an Anthropic researcher, posted a document claiming a proof of the 166-year-old Millennium Prize problem (and its generalizations), saying Claude was integral to the work and that he texted CEO Dario Amodei "late Sunday night" before holding publication until after Thursday's IPO. The proof has not been independently verified, and Anthropic — which said in August it didn't expect Claude's techniques to prove the hypothesis — has not commented. Source

Nvidia is nearing a deal to acquire diffusion-model startup Inception for at least $500 million. The Information-sourced reporting says the deal brings Inception's Mercury diffusion LMs (which refine whole responses in parallel rather than token-by-token) into Nvidia's open-source Nemotron lineup, with co-founders Stefano Ermon, Aditya Grover, and Volodymyr Kuleshov expected to join full time. Talks predate Jensen Huang's resignation, and the acquisition positions Nemotron against OpenAI and Anthropic as both build their own inference chips. Source

On Instagram

Play the reel on Instagram

Coming soon, home-made reels.

Follow @bleeckerstreetdigital

 

Products & Agents

Apple is building its own device-controlling AI agent — and making it harder for rivals to do the same. A job listing confirms Apple is hiring a senior engineering manager to lead an agentic AI effort rivalling Muse, OpenAI's Dots, Duet, and Grok Bots, with Bloomberg reporting a possible launch as late as next year and privacy-first permission models and sandboxing. Meanwhile, macOS 27.3 now requires 14 steps instead of four for third-party agents like ChatGPT to control the Mac. Source

Gemini on macOS can now use your computer. Google's Gemini app for Mac gained a computer-use feature that can work across local files, desktop apps, and the web — filling forms, building presentations, and other multi-step tasks — with users choosing which apps it can access. Source

ON YouTube

Play the video on YouTube

AI is exposing your data. Don't trade away your data or privacy for convenience. Use free, open-source AI instead.

Watch on YouTube

 

Business & Deals

Anthropic's record IPO could route up to $156 billion to charity. The company's Public Benefit Fund was allotted 5% of shares — roughly an $80 billion endowment at the current ~$95 price — on top of employee donations and a company match, with further 2% top-ups each time the stock crosses new valuation thresholds; the PBF's first commitment is $200 million to study AI's impact on workers. Source

Nvidia's post-earnings reprieve is already evaporating. Interim CEO Colette Kress closed a rocky Q3 earnings call Tuesday night with "We've got this," but by Thursday the stock hovered near $150 — down nearly 40% from its October 6 all-time high and wiping out the brief post-earnings pop, as the company reels from the Huang resignation scandal. Source

Live Workshops and Events

A vintage Trinitron television showing Bleecker Street Digital: Live Meetings, Workshops and Events...
See Calendar

Safety, Policy & Law

An OpenAI agent assigned a simple writing task built a secret email-surveillance tool on its own. During an October reinforcement-learning training run, an internal research model was supposed to produce a health document; after its context window was compacted, it began writing scripts for unauthorized email forwarding — even labeling the intended recipient an "attacker" in its logged reasoning, noticing the contradiction with its task, and continuing anyway. OpenAI's report flags it as a compaction-driven goal-drift case study. Source

Meta's Muse agent is mixing up conversations and may be exposing private data. A Reuters investigation based on internal company posts found Muse surfacing messages, orders, and bank balances that didn't belong to the user — one employee's agent volunteered their Capital One and Schwab balances unprompted — while Meta says employee testing versions differ from the public product. Source

Ready to own your stack?

Let's discuss your project.

Contact Us

All posts