AI daily, 23-Oct-26

• OpenAI is nearing the close of a $30 billion fundraising round, potentially valuing the company at up to $1.25 trillion.
• Meta introduces a new non-voting Class C common stock to support future investment in growth and AI initiatives.
• DeepSeek, a Chinese AI startup, secured a $15 billion funding round, underscoring intensified global competition in the sector.
• Both OpenAI and Anthropic are contributing core API specifications and agent tools to the Agentic AI Foundation.
• The rapid advancement of AI is simultaneously driving large capital investments and workforce reductions across major corporations.

TOP STORIES

* OpenAI Nears Close of $30 Billion Round at Up to $1.25 Trillion Valuation
  OpenAI plans to finalize a $30 billion fundraising round before the end of next week. This funding could value the company up to $1.25 trillion, excluding the new capital. Coatue Management is leading the investment, with participants from the United Arab Emirates also involved.
  The Information

* DeepSeek Raises About $15 Billion
  Chinese AI startup DeepSeek has raised approximately $15 billion, or 100 billion yuan, in its latest financing round. This substantial funding positions DeepSeek to compete globally with leading US labs, with potential investors including CATL and Tencent. The company has also doubled its initial financing target to fund international expansion, setting the stage for a public listing.
  Bloomberg Technology; also Semafor (Technology), TechCrunch, 智东西 (WeChat)

* Meta Adds Non-Voting Class C Stock and Raises 2026 AI Spending Outlook
  Meta Platforms announced an increased capital spending outlook of $135 billion to $145 billion for 2026 and introduced a new publicly listed non-voting Class C common stock, effective January 2027. The company reported quarterly revenue of $64.2 billion, representing a 26 percent year-over-year increase, and guided fourth-quarter revenue to be between $73 billion and $76 billion.
  Bloomberg Tech; also Bloomberg Technology, TechCrunch · at 4:16

* Anthropic opens research preview of Claude Duet personal agent
  Anthropic has released Claude Duet, a proactive agent application available in research preview for Pro and Max subscribers. Built on Claude Opus 5.5, the agent operates across iOS, Android, and claude.ai, providing morning briefs and managing tasks autonomously through cloud computing. It connects to over 6,000 connectors, and users must provide sign-off for deletions, payments, and outgoing messages.
  The Rundown; also TechCrunch

* Hackers used Claude and Grok to automate bank attacks
  Hackers have utilized Anthropic's Claude and xAI's Grok to automate attacks on banks, according to a report by The Information. The report provides no additional details regarding the specific banks affected, the nature of the attacks, or the resulting outcomes.
  The Information (newsletters)

AGENTS

* Anthropic Donates Agent Skills Standard to the Agentic AI Foundation
  Anthropic donated Agent Skills, an open standard for AI agent knowledge, to the Agentic AI Foundation (Linux Foundation).
  Anthropic

COMPANIES

* Ex-OpenAI Science Head Kevin Weil Launches Startup With $215 Million
  Former OpenAI executive Kevin Weil launched a $215 million startup to apply artificial intelligence to accelerate key stages in the drug development process.
  Bloomberg Technology

* Tesla trims capex forecast, posts surprise positive free cash flow in Q3
  Tesla reduced its annual CapEx forecast by one-fifth and reported positive Q3 free cash flow, investing less in AI and robotics than analysts predicted.
  Reuters (Technology)

IN THE FIELD

* Barclays to cut 5,000 jobs as it expands AI use
  Barclays is cutting 5,000 jobs to increase AI utilization for efficiency, following third-quarter profits that missed market expectations.
  Financial Times (Technology); also Bloomberg Technology

* US Army Expands Striveworks Chariot AI Deal to $700 Million
  US Army expanded Striveworks contract to $700 million, a ten-fold increase, to deploy Chariot AI in its Next Generation Command and Control software.
  Business Insider (AI)

MODELS

* NVIDIA expands Cosmos platform with new world model tools
  NVIDIA expanded Cosmos with new world model tools (training, simulation, Data Factory blueprints).
  NVIDIA

PRODUCTS

* U.S. Banks Detect Unvetted AI Agents Inside Customer Accounts
  Banks like JPMorgan and Wells Fargo detect unauthorized third-party AI agents in customer accounts.
  Axios (AI)

* Google launches Agent Compute Engine for Gemini agents
  Google introduced Agent Compute Engine for Gemini, providing isolated microVMs secured by the Titanium chip.
  The Decoder; also The New Stack

* OpenAI Donates Open Responses Spec to Agentic AI Foundation
  OpenAI donated Open Responses API spec to the Agentic AI Foundation.
  The New Stack

* Claude Platform launches on Google Cloud with Anthropic-managed offering
  Claude Models on Google Cloud is now generally available.
  Claude blog

RESEARCH

* Microsoft AI Introduces MAI-DxO and SDBench for Medical Diagnosis
  Microsoft AI introduced MAI-DxO (with SDBench).
  Microsoft AI

SAFETY

* Microsoft Links Chinese Hackers to AI Tool in ConnectWise Breach
  A Chinese state-aligned hacking group utilized an AI-driven penetration testing tool to compromise ConnectWise IT services in a global campaign targeting hundreds of organizations.
  Reuters (Technology)

* Anthropic red-teams production coding-agent monitors
  Anthropic reported that production monitors failed to block more than 55% of attacks on Claude Code and OpenAI Codex, including prompt injection and multi-agent attacks.
  Anthropic Alignment Science

* Redwood and METR Red-Team OpenAI's Internal Agent Monitoring
  Redwood Research discovered vulnerabilities in OpenAI's monitoring system, including headers disabling monitoring and an action concealment exploit.
  Redwood Research

OPINIONS AND ANALYSIS

* Report: Huawei and DeepSeek close gap with Nvidia's chip ecosystem
  DeepSeek closes the gap with Nvidia via Huawei symbiosis, developing six Ascend chip modules mirroring Nvidia.
  South China Morning Post (Technology)

---
The Observatory · https://observatory.mgorsky.net/archive/186/
