Meta launches Muse Wonder, claiming frontier-level capabilities as benchmarks and spending plans emerge. Google DeepMind ships Gemini 4 Argon with a safety-first rollout, matching GPT-6 Astra at lower cost. JPMorgan blocks outside AI agents from client accounts. OpenAI works through safety issues around loop transformers and weighs buying Trail of Bits. Character.ai files for Chapter 11, and regulators in the UK and Europe press for binding AI oversight.
Top stories
Meta launches Muse Wonder model
Meta launches Muse Wonder, a frontier-level model built for its Muse assistant, a milestone in chief AI officer Alexandr Wang's overhaul. Wang told Bloomberg how Meta positions it against OpenAI and Anthropic. Arena tests show one-shot 3D scenes, animated HTML and front-end designs versus Astra, Opus 5.5, GPT-6.1 Sol, Claude Sonnet 5.5, Fable, Kimi and Mistral, with costs listed. On Cheeky Pint, Zuckerberg and Wang described late-2025 talks on spending $100-200 billion on AI before the models proved themselves; CFO Susan Li warned Wang failure would be financially very bad.
Bloomberg Technology; also arena, Business Insider (AI), DHH and 1 more · details differ from the announcement
Google DeepMind releases Gemini 4 Argon with a safety-first rollout
Google DeepMind releases Gemini 4 Argon, its first proprietary model above Flash in over seven months, safety-first, built for coding, finance, legal, cybersecurity and science. Before wider release, cyber defenders got limited access via the Fairwind Program while Google scanned systems; Wiz used Argon to find a critical issue, cited as evidence the capabilities matter. On the Artificial Analysis Intelligence Index it matches GPT-6 Astra at 60% of the cost per task, discounted prices until November 30, with agentic gains and a 15% halving.
Google DeepMind; also Artificial Analysis
JPMorgan Blocks Outside AI Agents From Client Accounts
JPMorgan is blocking external AI agents, including Meta's Muse and Instinct, from accessing client accounts. CEO Jamie Dimon said on the bank's earnings call that it will not let any outside agent inside the company because it is too risky for clients and for JPMorgan.
The Information
OpenAI Works Through Safety Problems Before Releasing Next Flagship Model
OpenAI is working through safety problems tied to loop transformers, or recurrent depth, a controversial model design it has used, as it prepares its next unreleased flagship model for what leaders consider a safe release. The Information reported last month that the technique can hinder monitoring of how the AI reasons. Microsoft accidentally confirmed OpenAI's use of the technique in GPT-6.1 Sol last week, in a post since edited.
The Information
OpenAI in Early Talks to Acquire Trail of Bits
OpenAI is in early talks to acquire Trail of Bits, a startup that finds and fixes vulnerabilities in open-source software, according to people familiar with the matter. OpenAI is trying to build a substantial cybersecurity business. The deal size is unknown, though OpenAI has sometimes paid low hundreds of millions in stock for small startups. Talks continue and the deal could fall apart.
The Information
Companies
Character.ai Files for Chapter 11 Bankruptcy
Character.ai, whose users create and interact with AI personas, filed for Chapter 11 Monday, citing legal, regulatory and operational headwinds.
The Information; also Hacker News
Applied Intuition receives offers valuing it at $30 billion
Applied Intuition, which makes software to train self-driving vehicles, received offers valuing it at at least $30 billion, plus acquisition approaches, likely leading it to raise at least $1 billion in coming months, double its…
The Information (newsletters)
Microsoft restructures Azure cloud unit in AI-focused overhaul
Microsoft is restructuring its Azure cloud unit, cloud chief Scott Guthrie told staff Monday, consolidating units and reorienting around AI; Arun Ulag, who oversaw Azure databases, takes on additional responsibilities.
The Information
Decagon weighs acquisition interest from ServiceNow and others
Decagon, an AI enterprise customer-support agent company last valued at $4.5 billion, hired a bank to assess acquisition interest and held early talks with ServiceNow and others.
Business Insider (AI) · according to leaked documents, reported by Business Insider
Compute
Anthropic plans first Japan data center with NTT Data
Anthropic is in final talks with NTT Data to build its first Japan data center, Nikkei reports.
The Information (newsletters)
Nvidia to announce $2B in European investments
Nvidia CEO Jensen Huang plans to unveil $2 billion in new European investments next week, saying Europe needs to get in the race.
Semafor (Technology)
CtrlS plans $1.05bn AI factory in eastern India
India's CtrlS will invest $1.05 billion to build a hyperscale AI factory in eastern India.
Reuters Tech News (@ReutersTech)
Policy
OpenAI and Anthropic Back Binding AI Regulation at UK Parliamentary Hearing
At a UK parliamentary session, OpenAI and Anthropic backed a statutory AI-safety framework; Anthropic's Liam Booth-Smith called voluntary commitments insufficient, OpenAI's Tom Duff Gordon backed mandatory incident reporting.
Wired
Watchdogs Signal Tighter AI Oversight as US and China Hold Back
US and European banking regulators increasingly question how institutions use AI models, fearing threats to financial stability, Reuters reports.
Semafor (Technology)
Publishers form coalition against unauthorised AI use of content
The New York Times, Agence France-Presse and The Economist have joined over 40 other publishers in a coalition aiming to stop AI firms using or profiting from their content without permission, Reuters reports.
Reuters (Technology)
Trump administration plans to move AI testing agency staff to White House
The Trump administration plans to relocate officials from an AI testing agency to the White House, The Information reports, shifting staff into the executive office.
The Information (newsletters)
Products
OpenAI leads Anthropic in agent platform use, VentureBeat survey finds
VentureBeat Intelligence's September 2026 survey found twice as many companies use OpenAI's agent orchestration platform as Anthropic's.
VentureBeat (AI)
Safety
Anthropic says defenses cut Claude reasoning-token extraction by up to 99%
Anthropic says new defenses cut extraction of Claude's internal reasoning tokens by 90-99%.
The Decoder; also Anthropic
Anthropic warns small AI models could soon self-replicate on IoT devices
Anthropic says small AI models may be months from self-replicating on commercial IoT devices, including smart ovens and off-brand Android TV boxes.
Reuters (Technology); also Financial Times (Technology), Semafor (Technology)
Greg Brockman says GPT-5.6 Sol found a critical Chrome vulnerability
Greg Brockman says six days and $40,000 of GPT-5.6 Sol compute found a critical Chrome vulnerability, earning Google's first full-chain reward in 2026 for a flaw affecting an estimated 1 billion users.
Greg Brockman (@gdb)
Sent by the Observatory.