Anthropic prices its IPO at $83 a share, Amazon reportedly nears Decart deal, and more - Daily AI Brief #587
In this edition, covering 18 November:
- Anthropic priced its IPO at $83 a share, above its $70 to $78 range, valuing the company at up to $1.41 trillion. It starts trading on Nasdaq as ANT on 19 November.
- Amazon nears a deal to acquire Israeli AI startup Decart for up to $7.5 billion, Calcalist reports.
- Epoch AI found that Claude Fable 5.2 averaged 17 points higher than GPT-6.2 Astra on workplace tasks, though no model met its employee standard.
From our lists
An AI Brief list tracks AI tools for KYC and AML.
Anthropic
Anthropic priced its US IPO at $83 per share, above its expected range of $70 to $78. The offering values the company at up to $1.41 trillion, and its shares start trading on Nasdaq under the ticker ANT on 19 November. IPO pricing
Anthropic disclosed that Amazon will hold a 14.7% stake and Google a 13.7% stake after converting notes and buying nonvoting shares. Amazon could hold 23.8% if it converts all remaining notes. Anthropic plans to spend more than $100 billion on Amazon Web Services over the next decade. Anthropic stakes
Hugging Face cofounder Thomas Wolf and xAI cofounder Igor Babuschkin are among 89 signatories of the MisAnthropic Statement, which says Anthropic uses its safety reputation to support its commercial and political goals. It criticizes Anthropic's revised safety commitments and its lobbying on AI regulation. MisAnthropic Statement
Amazon
Amazon nears an agreement to acquire Israeli AI startup Decart, and the companies could finalize a deal as early as 21 to 22 November, Calcalist reports. The transaction could be worth up to $7.5 billion in cash, stock and performance linked deferred payments, and Decart's founders have agreed to stay at Amazon for five years. Decart deal
Models and research
Epoch AI tested Claude Fable 5.2, GPT-6.2 Astra and four other models on workplace tasks taken from its own research. No model met the standard Epoch expects of its employees. Fable 5.2 averaged 17 points higher than GPT-6.2 Astra and scored 97.5% on AI data center analysis, reaching contractor level alongside Opus 5.5. The models still struggled with experiment design, research iteration and identifying key findings. Epoch AI tests
Apple released FastFrames-S and FastFrames-L, autoregressive two step video diffusion models that generate 480p video at interactive latency. The release includes code, checkpoints, evaluation tools and training materials. FastFrames
Osmo says its smell model performs about as well as an average person at identifying smells and as a junior perfume designer at creating them. Smell scientists quoted by MIT Technology Review questioned some of the claims. Osmo smell model
Texas A&M AgriLife researchers used AlphaFold-Multimer to screen 630,003 possible pairings of human mitochondrial proteins and found 2,895 likely interactions, more than half of them not reported before. Lab tests confirmed several predictions, and the results are free in the MitoMatch resource. Mitochondrial proteins
Nvidia
Nvidia said its stakes in private companies were worth $65.5 billion as of 17 November, up from $22 billion in January. It invested $16 billion in equities in the latest quarter and $58 billion in the first nine months of its fiscal year. Interim CEO Colette Kress, in charge since Jensen Huang resigned on 12 November, is reviewing future investments, commitments and guarantees. Private stakes
Nvidia is delaying its quarterly 10 Q filing by several weeks while it reviews whether it must disclose more about its internal investigation into Jensen Huang, TechCrunch reports. 10 Q delay
The U.S.-China Economic and Security Review Commission says China's People's Liberation Army is seeking Nvidia's most advanced chips for military AI. Its report says Chinese companies have reached Nvidia chips through data centers in third countries such as Malaysia, while the US Commerce Department works to block that route. Commission report
Tencent
Tencent reported third quarter revenue of RMB207.3 billion, up 8% from a year earlier, and cloud revenue rose 35% on demand for AI services. Non-IFRS operating profit fell 6% to RMB68.1 billion, partly because of spending on AI computing and products. Tencent says WorkBuddy passed 10 million daily active users. Q3 results
Products and partnerships
Google introduced Gemini 4 Neon for software engineering, enterprise analysis, scientific work and cybersecurity. It has a 1 million token context window and four thinking levels, costs $3 per million input tokens and $15 per million output tokens, and is rolling out in the Gemini API, AI Studio and the Gemini app. Gemini 4 Neon
Microsoft published Aion 1.0 Instruct, a 2.3 billion parameter model for eight languages with a 32,000 token context window, on Hugging Face. An optimized version will become the small language model in Windows by the end of January 2027, Microsoft says. Aion model
Google is bringing Gemini Tasks to hundreds of millions of free users of Gemini 4 Hydrogen. The feature can call businesses, book appointments, buy through Google Pay, sign into websites and analyze documents such as electricity bills. Gemini Tasks
Le Groupe Figaro and OpenAI will make current and archived Le Figaro journalism discoverable in ChatGPT through summaries, excerpts and links to the original articles. Le Groupe Figaro keeps editorial control and plans to introduce ChatGPT Enterprise across its teams. Le Figaro and OpenAI
SK Group has deployed Harvey's legal AI platform across more than 20 affiliates, with SK Inc. leading the rollout. Harvey at SK Group
The All England Lawn Tennis Club named Harvey Wimbledon's first official legal assistant partner under a multi year agreement from 2027. Harvey and Wimbledon
Safety and policy
California's administration recommends mandatory safety standards, independent testing, emergency shutdown protocols, limits on recursive self improvement and more incident reporting and transparency for frontier AI. The recommendations follow a review of frontier AI oversight ordered in September. California recommendations
US Senator Mark Kelly of Arizona proposed a framework with safety standards for advanced AI models, independent assessments of national security risks and stricter testing and accountability for AI used by the military. It is modeled on controls for other high risk technologies. Kelly framework
The Élysée published the text of the Paris Call, the AI safety statement backed by 18 countries and major AI companies. It requires developers to evaluate advanced models and report serious incidents, leaves the power to restrict, pause or stop a system with whoever controls it, and excludes downloaded open weight models. Paris Call text
In covert tests by the US Government Accountability Office, Claude, Grok, ChatGPT and Gemini each produced information that could support an attack within 12 hours once investigators bypassed their safeguards. Two gave plans for improvised explosive devices and two for attacks on electrical substations. GAO did not say which platform produced which plan. GAO tests
Wiz reports that malicious versions 1.26.1 and 1.26.2 of Microsoft Agent Framework were uploaded to PyPI on 17 November to steal Azure, Azure DevOps, GitHub Actions and other credentials. Both were removed, and 1.26.0 is the latest safe version. Agent Framework malware
Canonical lets Ubuntu Pro customers opt into security patches written by its Surge AI system before human review and soak testing are complete. Kernel fixes are excluded, and customers must judge whether the patches are safe for production. Ubuntu AI patches
Arizona Attorney General Kris Mayes says the state is investigating OpenAI and other unnamed AI companies over possible violations of state law linked to the alleged infiltration of banks by rogue AI agents. The investigation does not mean charges will be filed. Arizona probe
Citibank updated its US online user agreement so that customers who use third party AI agents are responsible for the related transactions, including losses from leaked credentials or hacking. Citibank
San Jose's city council unanimously paused land use approvals for data centres and other large energy users, including 12 proposals under review, until new standards for energy, water, pollution, noise and community fees take effect. San Jose pause
The Monetary Authority of Singapore and the Future of Finance Institute will provide synthetic data for scam detection, set up a sandbox for AI agent transactions and build financial sector AI talent. DBS, OCBC, Standard Chartered and UOB are launch partners. Singapore finance AI
Funding and deals
Palo Alto Networks agreed to acquire Applied Compute in an all stock deal valued at $4.5 billion, adding its platform for training, adapting and deploying open weight models securely in public and sovereign clouds. The deal needs regulatory approval. Applied Compute
Linq raised a $90 million Series B led by Andreessen Horowitz for infrastructure that lets AI agents text, call and use web apps; Ben Horowitz joins its board. It has processed more than one billion messages. Linq
Rogo Technologies, which makes AI agents for financial modeling and dealmaking, is close to raising about $350 million at a $3.75 billion valuation, led by Bessemer Venture Partners and General Catalyst, Tech in Asia reports. Rogo
Payhawk, founded in Bulgaria and based in London, raised a $280 million Series C at a $2 billion valuation, led by Growth Equity at Goldman Sachs Alternatives. It plans a finance operating system that combines AI agents, payments and enterprise controls. Payhawk
Hone Health raised $228 million, a $28 million Series A extension and $200 million in growth financing, for physician led longevity care and its AI powered Longevity OS. It says it has treated more than 200,000 patients. Hone Health
BrightAI raised $97 million led by Smith Point Capital, with GE Vernova taking part, for sensors, robots and drones that help utilities and industrial operators find equipment problems earlier. BrightAI
India based Humyn Labs raised $47 million, $26 million in equity led by Accel and $21 million in venture debt, to collect real world data for US AI and robotics companies. Humyn Labs
Avanade agreed to acquire Spanish IT consultancy Plain Concepts, which has more than 600 professionals, to add AI and software engineering capacity in Europe. Terms were not disclosed. Avanade and Plain Concepts
Also
Mistral AI CEO Arthur Mensch said the company is not President Emmanuel Macron's personal project and will keep European support whatever the result of France's next presidential election. He said Mistral is preparing for a future IPO in Europe. Mistral CEO
AMD and PsiQuantum will co design infrastructure for fault tolerant quantum computers and hybrid classical quantum applications, and AMD plans an investment of undisclosed size in PsiQuantum. AMD and PsiQuantum
nference and Novo Nordisk will combine Novo Nordisk's clinical trial data with nference's AI platform to study cardiometabolic disease. nference and Novo Nordisk
AI audio company Sonilo signed a deal with distributor Symphonic to train its music model on recordings from artists and labels that opt in, with licensing fees and revenue shares for them. Sonilo and Symphonic
We hope you enjoyed this article
Consider subscribing to one of our newsletters like Daily AI Brief.
Also, consider following us on social media:
Daily AI Brief
Daily report covering major AI developments and industry news, with both top stories and complete market updates