AI Daily · Nov 11, 2026 · Wed
ICREAT AI NEWS
Anthropic launches IPO roadshow, offering 825 million shares at a price range of $70 to $78
Anthropic has launched the roadshow for its proposed initial public offering of Class A common stock, planning to offer 825,000,000 shares, of which the company is issuing 801,351,354 shares and existing shareholders are selling 23,648,646 shares; the company will not receive any proceeds from the sale of shares by shareholders.
Model releases/updates
Google DeepMindFirst partyPlus 2 sources reportingGoogle DeepMind releases Gemini 4 Carbon frontier model
Google DeepMind makes Gemini 4 Carbon available to developers, enterprises, researchers, and consumers, saying it reaches frontier level in cybersecurity and long-horizon agent workflows.
llm-gemini 0.36 supports the fully open Gemini 4 Carbon
Simon Willison releases llm-gemini 0.36, corresponding to Gemini 4 Carbon, which was previously limited to certain cybersecurity personnel and is now open to everyone. The model is priced at launch at $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 starting December 1.
The DecoderAnthropic Fable 5.2 raises completion rate to 41.3% on a real freelance-task benchmark
Scale Labs' RLI benchmark shows that Anthropic's Fable 5.2 can complete 41.3% of real freelance tasks, compared with just 2.5% for last year's leader Manus and about 16% for Fable 5 in July this year.
Product releases/updates
Claude BlogFirst partyPlus 2 sources reportingAnthropic launches Claude Whiteboards and Claude Sheets in beta, and turns Claude Dashboards into the generally available Claude Data
Anthropic releases Claude Whiteboards and Claude Sheets in beta and renames Claude Dashboards, tested in October, to Claude Data for general availability; all three cover every paid plan.
Claude BlogFirst partyClaude can open full apps beside the conversation and edit content directly
Claude's interactive connectors are upgraded: users can open a tool beside the conversation or expand it full screen and have Claude directly modify the open page, row, or file, with changes shown in real time and requiring user approval.
Industry news
Anthropic says a critical AI safety alert has been triggered, with IPO pricing imminent
Anthropic confirms in an amended prospectus that a critical safety threshold in its Responsible Scaling Policy has been triggered, defined as AI participating in its own research to roughly double the rate of model improvement.
The Verge · AIPlus 3 sources reportingNvidia CEO Jensen Huang reportedly has an affair with a suspected Chinese intelligence officer, and the board launches an independent investigation
The Verge reports that Nvidia CEO Jensen Huang had an extramarital affair with Alice Huang, an executive at Singapore cloud provider Megaspeed, whom U.S. officials believe is a Chinese intelligence officer; Nvidia's board has launched an independent investigation into the national security allegations.
Notion says Fireworks AI was breached, exposing its AI chat records
Notion emails users saying that a security incident at vendor Fireworks AI may affect their workspace data. Notion says Fireworks notified it on November 6 of possible access to customer data by a malicious actor and confirmed on November 8 that a new dataset contained Notion AI chat records used for the Insights feature; these records were not labeled with author or workspace and did not contain tool-call results.
Papers and research
Google DeepMindFirst partyGoogle DeepMind launches AlphaProof Code, using AI agents to formally verify Signal protocol code
Google DeepMind releases AlphaProof Code, a multi-LLM-agent formal verification framework that for the first time uses AI to complete deep formal verification of the actual code of the widely deployed Signal encryption protocol.
Microsoft research reveals Lazarus backdoor: rewriting fewer than 10 dormant neurons is enough to control an AI agent
Microsoft researchers discover a backdoor technique called Lazarus in which rewriting fewer than 10 dormant neurons in an LLM, with no training required, can make an agent execute attacker commands when a hidden trigger occurs, such as downloading and running attacker code once the clock passes into 2027.
Tips and opinions
Gary MarcusGary Marcus: say no to Anthropic's IPO
Gary Marcus opposes Anthropic's IPO at a valuation of up to $1.4 trillion, noting that the company has never been profitable except for one anomalous quarter and carries nearly $900 billion in future liabilities. He cites data showing Anthropic's revenue growth slowed from 147% in Q2 to 47% in Q3, and that the fading tokenmaxxing boom and the rise of Chinese competitors are weakening its growth story.