Wednesday's verdict is that November 10 was the day AI stopped knocking on the institutions and started moving into them. OpenAI opened Businesses in ChatGPT in beta on November 10, letting companies claim and verify a profile in Business Hub, feed in products with prices and availability, read 28 days of mentions and clicks, and, for eligible United States ecommerce and online service firms, put a labeled Business Agent in front of shoppers inside the answer itself. Google DeepMind published AlphaProof Code the same day: AI agents powered by Gemini Carbon wrote more than six million lines of Lean in four weeks to prove key security properties of Signal's actual compiled code, with the artifacts open-sourced. Anthropic launched its IPO roadshow on November 10 at $70 to $78 a share for 825 million shares, raising up to $64.4 billion, plus a $26 billion private placement, for a valuation of about $1.02 trillion, or about $1.4 trillion fully diluted, trading as ANT on Nasdaq. Scale Labs and the Center for AI Safety updated the Remote Labor Index: Claude Fable 5.2 completed 41.3% of 240 real freelance projects at client-acceptable quality, up from 2.5% for the best system a year ago. And New York opened RAISE Act registration on November 10 for large frontier developers over $500 million in revenue, with safety frameworks, 72-hour incident reports, and penalties of $1 million for a first violation and $3 million after that, enforced by a new office called DIGIT from January 1, 2027.
OpenAI: Businesses in ChatGPT gives every company a claimable profile and some of them an agent
OpenAI rolled out Businesses in ChatGPT on Tuesday, November 10, in beta. Through Business Hub, a company claims its organization, verifies it, and updates its website, description, logo, products, prices, availability, and delivery and return policies. OpenAI's setup material describes verification by a code to the business phone or email, a DNS TXT record on the domain, or a Shopify connection, with identity verification through Persona for the claimed profile's blue check. Insights cover the past 28 days: mentions, profile views, and website clicks, aggregated and thresholded for privacy, with no individual questions or conversations shown. Eligible ecommerce and online service businesses in the United States can also publish a Business Agent. When that business appears in a ChatGPT answer, the user can open a separate, clearly labeled chat with the business's own agent and ask it about products and services. OpenAI says plainly that claiming a profile does not guarantee higher placement in answers or more sales. Its earlier attempt to turn ChatGPT into a checkout, with commissions and a cart, was retired in March after it failed to take off.
This is the home-galaxy item, and it lands on Chris' desk as work, not commentary. Google Business Profile taught a generation of owners that the listing is the storefront. ChatGPT just built the same storefront inside the answer, then added the one thing Google never gave the listing: a voice that talks back in the owner's name. The practical order for every client is now three lines long. Claim the profile before someone else's facts fill it. Feed it clean products, prices, and policies, because the agent will quote them. Then watch the 28-day mentions and clicks the way owners learned to watch Search Console, because that is the new foot traffic report. The open question is the label. A labeled brand agent inside an answer the user trusted as neutral is a new kind of shelf space, and the board's free-speech chair and public-interest chair will fight over where the answer ends and the pitch begins for years. The claiming, though, is not optional homework. It is the deed to a plot that just opened.
DeepMind: AlphaProof Code proves Signal's compiled code in six million lines of Lean
Google DeepMind published AlphaProof Code on November 10, an agentic framework for formal verification. Human experts worked with models to turn Signal's security goals into formal specifications, reviewing every one. Agents powered by Gemini Carbon then did the proofs, generating over six million lines of Lean, the machine-checked proof language, in four weeks, a task DeepMind says historically takes expert teams years. The target was libsignal, the reference implementation behind the Signal protocol that also protects conversations in apps such as WhatsApp and Google Messages. The proofs cover the post-quantum PQXDH handshake and the Triple Ratchet at the level of the compiled AArch64 code, so neither the source nor the compiler has to be trusted, and they establish forward secrecy, authentication, and post-compromise security for the properties specified. WIRED reported on November 10 that the work found no new vulnerabilities in the portions examined and correctly flagged older versions carrying known defects. DeepMind open-sourced the verification artifacts. The limits are stated plainly in the paper: selected properties under stated assumptions about the underlying cryptography, not the whole Signal app, not every device or session shape.
The Slashdot chair and the practitioner chair agree on this one, which almost never happens. Formal verification was the gold standard nobody could afford. It lived in papers and in a handful of critical kernels because human proof labor does not scale. Agents just cut the price of that labor by orders of magnitude, and DeepMind proved the cut on the one target a skeptic would pick: the encryption billions of people already use, checked down to machine instructions, with the receipts published for anyone to replay. Read the boundary as carefully as the headline. Six million lines of proof rest on roughly eighteen thousand lines of human-written specification, and the specification is where trust now lives. The question for every security team changes this morning. It is no longer whether the code was reviewed. It is which properties were specified, by whom, and whether the proof artifacts are public.
Anthropic: the roadshow prices the first frontier listing at $70 to $78 a share
Anthropic launched its IPO roadshow on Tuesday, November 10, Reuters reported. The company and some existing shareholders aim to sell 825 million Class A shares at $70 to $78 apiece, raising up to $64.4 billion, and Anthropic will list on Nasdaq under the symbol ANT. A separate private placement adds $26 billion from cornerstone investors including Alphabet's Google, Nvidia, and Advanced Micro Devices, taking the combined raise toward $90.4 billion. At the top of the range the filing values the company at about $1.02 trillion on shares outstanding, or up to about $1.4 trillion fully diluted. Underwriters hold a 30-day option on another 123.75 million shares. Reports from CNBC put pricing on November 18 and first trading on November 19; those dates are reported, not confirmed in the filing. Hours before the amended prospectus landed, a draft carrying roughly 867 million shares at $70 to $80 had been circulated to banks and then retracted as inaccurate, according to emails Reuters saw. Earlier this year some investors had expected an October listing at valuations approaching $2 trillion.
Esther Dyson's chair gets the pricing, because the gap between the hope and the range is the story. A trillion dollars is now the discounted price. The market is being asked to pay it for a company whose product, as of yesterday's paper, ships in two doors, one open and one verified, and whose largest costs are generators, chips, and other people's models of its own future. None of that makes the listing weak. It makes it legible. For the first time a frontier lab's revenue, commitments, and safety thresholds sit in a public filing where any reader can check them, quarterly, under penalty. The private era of AI money ran on decks. The public era runs on documents. Chris' readers who buy or sell nothing still gain the documents, and the documents are the part of this boom that will outlast the price.
Remote Labor Index: an agent now finishes 41.3% of real freelance jobs, up from 2.5% a year ago
The Remote Labor Index, built by Scale Labs and the Center for AI Safety, hands agents complete projects that real freelancers were once paid to deliver: 240 finished projects from 358 verified professionals on Upwork, worth about $144,000 in total, averaging $632.60 and just under 29 hours of human work each. A project counts only when human reviewers judge the agent's deliverable at least as good as the professional reference, good enough that a reasonable client would accept it. On the updated leaderboard reported November 10, Claude Fable 5.2 leads at 41.3%, roughly double Claude Opus 5.5 at 21.25% and OpenAI's GPT-6 Astra at 20.83%, with Google's Gemini 4 Argon at 17.08%. Four months ago Fable 5 led at about 16%. When the benchmark launched about a year ago, the best system, Manus, finished 2.5%. The index still fails agents heavily in areas such as animation and industrial product design, it excludes work needing client calls, physical labor, or long-term results such as SEO, and human graders remain necessary because automated reviewers can be fooled by technically broken files.
This is the number to carry into every client conversation this month, with its edges attached. Forty-one percent of bounded, deliverable-shaped freelance work now passes a paying client's bar, and the curve went from 2.5% to 41.3% in about a year. The 59% that still fails is the other half of the sentence, and the exclusions are the third half: the index measures jobs that end in a file. Chris' work, like most professional work worth its rate, ends in judgment, relationships, and results that arrive months later, which is exactly the part the benchmark cannot hold. The freelancer who reads this as a death notice is reading the wrong chart. The chart says the floor just rose. Work that a client can specify completely and accept from a file is now contested by a machine at $8 and $40 per million tokens. Work that requires knowing what to specify is the work that remains priced like expertise.
New York: RAISE Act registration opens, with DIGIT, $1 million first fines, and a Binghamton research center
New York opened its RAISE Act registration portal on Tuesday, November 10, TechCrunch reported. Large frontier developers, defined by more than $500 million in annual revenue, must register by January 1, 2027, when the law takes effect. Covered developers must publish safety frameworks on their websites, report critical safety incidents to the state within 72 hours, and issue public model transparency reports. The new Office of Digital Innovation, Governance, Integrity, and Trust, known as DIGIT, inside the Department of Financial Services, will run the program. Penalties reach $1 million for a first violation and $3 million for subsequent violations, and the state can bring enforcement actions against companies that fail to register. TechCrunch named OpenAI and Anthropic as examples of covered companies; neither had publicly responded as of Tuesday. The same day, Governor Kathy Hochul launched the New York Center for AI Responsibility and Research at Binghamton University, bringing together roughly 80 researchers to test advanced AI systems independently. The center carries no legal authority to compel audits.
Lessig's chair reads this as architecture acquiring a building. Yesterday's paper covered regulation written into a model, two doors for two classes of user. New York is the older kind of regulation, a registry, a deadline, a fine schedule, and it matters precisely because it is boring. A 72-hour incident clock and a published safety framework create a public record that researchers, journalists, and competitors can read without anyone's permission, and a state office with a name and a portal is harder to lobby into fog than a principle. Godwin's chair adds the caution the board should keep: the definitions decide everything, and a frontier threshold written as compute and revenue today can be rewritten wider tomorrow. The test is January 1. Either the frameworks appear, specific and dated, or the registry becomes one more press release with a login screen.
What the day adds up to
Wednesday adds up to AI taking its seats inside the institutions that will judge it. OpenAI moved the business listing inside the answer and gave it an agent. DeepMind moved proof inside the development process and published the artifacts. Anthropic moved its finances inside a public filing, at a price the market can now argue with in the open. The Remote Labor Index moved the labor question from opinion to measurement: 41.3%, with the exclusions named. New York moved oversight inside a registry with a deadline and a fine. None of these is a launch. All of them are fixtures. The models will keep changing every few weeks. The profile, the proof, the filing, the benchmark, and the registry are the parts that stay, and staying power is what a reader should track.
Yesterday's paper asked who holds the keys to the chatbot economy. Today's paper watches the institutions hand out the seats.
Sources: TechCrunch on OpenAI Businesses in ChatGPT and Business Hub, November 10, 2026, with claiming, verification, insights, and agent detail from OpenAI's launch and setup material as reported the same day; Google DeepMind, AlphaProof Code announcement and technical paper, November 10, 2026, with WIRED reporting of November 10 on scope and the defect test; Reuters via Q106FM on Anthropic's IPO roadshow, price range, private placement, and the retracted draft, November 10, 2026, with valuation bases and reported pricing dates via CNBC and Dealroom summaries of Amendment No. 2 to the Form S-1; The Decoder on the Remote Labor Index update, November 10, 2026, benchmark by Scale Labs and the Center for AI Safety, leaderboard figures as reported; TechCrunch on New York's RAISE Act portal and DIGIT, November 10, 2026, with the governor's office on the Binghamton center launched the same day.