AI News November 11, 2026: Anthropic IPO, DeepMind Proves Signal Secure, Gemini Robotics 2.5
November 11, 2026 was a day when AI news came from every direction at once: a major AI lab filed to go public, Google DeepMind used AI to mathematically prove the code behind encrypted messaging is secure, humanoid robots went to work on live car factory lines, and OpenAI bought its way into travel. Here is what actually happened — and what it means for you.
The short answer: Anthropic launched the roadshow for its initial public offering on November 10, offering 825 million Class A shares at $70–$78 each. Google DeepMind's AlphaProof Code formally verified key security properties of Signal's cryptographic protocol in four weeks, and its Gemini Robotics 2.5 model ran Apptronik Apollo 2 humanoids on Mercedes-Benz production lines with a 96% success rate. OpenAI launched Businesses in ChatGPT in beta and acquired travel startup Odessa, while Meituan released a 1.6-trillion-parameter open-weight model under an MIT license.
1. Anthropic Launches Roadshow for Its Proposed IPO
Anthropic, the company behind Claude, launched the roadshow for its proposed initial public offering on November 10, 2026. The offering covers 825 million Class A shares with an estimated price range of $70 to $78 per share, and the company has filed for a proposed listing on the Nasdaq. If priced at the top of the range, the deal would value the offering at roughly $64 billion in new shares sold — one of the largest tech listings of the year.
Why it matters: Anthropic is the first of the big frontier AI labs to test the public markets. The roadshow is the step where the company pitches institutional investors before pricing the shares, so the final price — and how much the market is willing to pay for a pure-play AI lab — will be the number to watch in the coming days. For users, an IPO mostly means more scrutiny of Anthropic's finances and more pressure to turn Claude's popularity into profit. Our earlier coverage of the AI landscape is in our AI news roundup for November 9.
2. DeepMind's AlphaProof Code Mathematically Verifies Signal's Encryption
Google DeepMind announced on November 10 that its new AlphaProof Code framework formally verified key security properties of the code behind the Signal protocol — the cryptography that protects billions of private conversations in Signal, WhatsApp and Google Messages. According to DeepMind, this is the first time AI has formally verified a widely deployed cryptographic protocol at this depth.
The numbers are striking. Human experts worked with AI agents to turn Signal's security goals into formal mathematical specifications, and agents powered by Gemini Carbon then generated more than six million lines of proof in Lean, a language that mechanically checks whether code does what it claims. The whole process took about four weeks — work DeepMind says historically takes expert teams years. To check the system could tell good code from bad, the team reintroduced two real, previously fixed Signal bugs into an isolated copy of the code; in both cases AlphaProof Code produced a machine-checked proof that the buggy version was insecure.
The verified properties include the post-quantum PQXDH handshake and the Triple Ratchet, covering forward secrecy, authentication and post-compromise security. DeepMind says it found no new vulnerabilities in the parts it examined and is open-sourcing the specification and proofs. The honest caveat, noted by DeepMind itself and independent analysts: the proof covers selected cryptographic guarantees of the protocol's implementation, not the whole Signal app, and it assumes the underlying cryptography is perfect. Still, connecting protocol-level guarantees to compiled machine code — down to the AArch64 binary — is a genuine milestone for software security, and it hints at a future where AI agents routinely audit the code that runs our most sensitive systems.
3. Gemini Robotics 2.5 Puts Humanoids to Work at Mercedes-Benz
Google DeepMind also introduced Gemini Robotics 2.5 on November 10 — its most capable vision-language-action model to date — and it is already working outside the lab. In a proof of concept on active Mercedes-Benz production lines in Kecskemét, Hungary and Sindelfingen, Germany, Apptronik's Apollo 2 humanoid robots performed automotive kitting — sorting individual metal parts from bulk bins into sequencing trays in the exact order each vehicle on the line needs — with a 96% success rate.
The system works as a pair: the Gemini Robotics ER embodied-reasoning model tracks the workflow, while the Gemini Robotics 2.5 actions model controls the robot's hands and arms. DeepMind says the model was tested on more than 200 tasks, from serving food with tongs to pipetting liquids, and that it completes tasks 20–30% faster than its predecessor. The release also includes ASIMOV-Jailbreak, an open-source benchmark for testing whether robots can be tricked into unsafe actions — a safety tool DeepMind is releasing publicly. Gemini Robotics 2.5 is rolling out to DeepMind's robotics partners, including Apptronik, Agile Robots and Boston Dynamics; there is no public API or price yet, so this is a partner-only rollout for now. A pilot is not a deployment, but a 96% success rate on a line that cannot pause is the strongest factory result a humanoid has posted so far.
4. OpenAI Launches Businesses in ChatGPT and Buys Travel Startup Odessa
OpenAI introduced Businesses in ChatGPT in beta on November 10. The feature lets businesses claim their profiles, manage the product information ChatGPT shows about them, and offer separate Business Agent conversations that turn interest into action — essentially a governed channel for commercial discovery inside ChatGPT. Eligible US ecommerce and online-service firms can take part first. For small businesses, this is worth watching: if customers start finding and buying through ChatGPT, the businesses that set up their profiles early will have the advantage. Our guide to the best AI tools for small business owners covers the wider stack this fits into.
OpenAI also made a move into travel. On November 11, the company acquired Odessa, a travel startup founded by Francis Davidson, who previously founded the hospitality company Sonder. OpenAI has not disclosed what it plans to do with Odessa or how the acquisition fits into a travel strategy, and no terms were announced. The deal is notable because it signals OpenAI may be exploring travel as a next frontier for its agents — booking trips is exactly the kind of multi-step task AI assistants are being built to handle.
5. Meituan Releases LongCat-2.5, a 1.6-Trillion-Parameter Open Model
Chinese tech giant Meituan released LongCat-2.5, a mixture-of-experts model with about 1.6 trillion total parameters and around 48 billion active parameters per token, under an MIT license. The open license is the story here: a model of this scale being released with permissive terms gives developers and researchers unusually large-scale agent-model weights to build on. It is the latest sign the open-weight race is accelerating alongside the closed frontier labs — and it arrived the same week Anthropic moved toward the public markets.
6. Also Today: New York's AI Safety Law Takes Effect, Zoom Offers $150K in AI Credits
New York has begun enforcing a new AI safety law that puts OpenAI, Anthropic and other major AI developers under formal oversight — one of the first state-level regulatory regimes for frontier AI labs in the US. The law adds a new compliance layer for the biggest model builders and will be closely watched by other states considering similar rules.
Zoom announced AI Launchpad on November 10, a grant program offering eligible startups up to $150,000 in complimentary credits to build AI-native apps on Zoom's speech, language and voice AI services, with no equity stake required. Applications opened the same day. And Google gave cyber defenders first access to Gemini 4 Carbon, its strongest model, which launched November 10 with frontier performance in cybersecurity and long-horizon agentic work, priced at $2/$10 per million tokens until November 30, 2026.
What to Watch Next
Three things to follow from today. First, Anthropic's IPO pricing — the final share price will tell us how public markets value a frontier AI lab. Second, whether DeepMind's formal-verification approach spreads beyond Signal: if AlphaProof Code can be pointed at other critical codebases, AI-assisted security auditing could become a real product category. Third, whether Gemini Robotics 2.5 moves from pilot to permanent deployment on factory lines — the 96% number has to hold at scale. We will cover each as it develops; catch up on yesterday's stories in our AI news roundup for November 8.
Frequently Asked Questions
What is Anthropic's IPO price range?
Anthropic's proposed IPO covers 825 million Class A shares at an estimated $70–$78 per share, with a proposed Nasdaq listing. The roadshow launched November 10, 2026; the final price will be set by investor meetings.
Did DeepMind really prove Signal is secure?
DeepMind's AlphaProof Code formally verified key security properties of Signal's cryptographic protocol implementation — including the post-quantum PQXDH handshake and Triple Ratchet — using machine-checked proofs in Lean. It found no new flaws in the parts examined. The proof covers selected guarantees, not the entire Signal app, and assumes the underlying cryptography is perfect.
What is Gemini Robotics 2.5?
Gemini Robotics 2.5 is Google DeepMind's latest vision-language-action model for robots. It connects visual and language understanding to robot movement, and in a Mercedes-Benz factory pilot it ran Apptronik Apollo 2 humanoids sorting car parts with a 96% success rate. It is rolling out to DeepMind's robotics partners only, with no public API yet.
What is Businesses in ChatGPT?
Businesses in ChatGPT is a beta feature from OpenAI that lets businesses claim profiles, manage how their products appear in ChatGPT, and offer Business Agent conversations that turn interest into action. It is aimed first at eligible US ecommerce and online-service companies.
What is Meituan's LongCat-2.5?
LongCat-2.5 is a 1.6-trillion-parameter mixture-of-experts model released by Meituan under an MIT license, with about 48 billion active parameters per token. The permissive license makes it one of the largest open-weight agent models available.
Is New York's AI safety law now in effect?
Yes. New York began enforcing its AI safety law on November 11, 2026, placing OpenAI, Anthropic and other major AI developers under formal oversight — one of the first state-level regimes of its kind in the US.
Related reading:
Comments
Post a Comment