Anthropic files to go public on Nasdaq

Anthropic filed to go public on Nasdaq, reporting $16.3 billion in first-half revenue, while SpaceX secured roughly $40 billion to lease AI chips. Also inside: Google DeepMind's Genie 4, OpenAI's claimed science results, new Siri features and robotics funding that more than doubled.

AI lab releases

Google DeepMind unveils Genie 4 world model

Google DeepMind introduced Genie 4, a world model, meaning AI that generates virtual environments, which users can steer in real time. It can now produce matching audio and lets people change a scene with typed prompts, such as altering the weather. The model is launching as a research preview for selected testers before reaching Google AI Ultra subscribers.

Why it matters: Tools that build playable worlds on demand give a window into how quickly simulated environments are maturing.

Sources: Google DeepMind, TechCrunch, The Verge, Ars Technica

Liquid AI releases small LFM3 model family

Liquid AI released LFM3, a family of small models around 2 billion parameters covering text, vision, audio and decision-making. The company says each model fine-tunes on a single GPU and runs on edge devices such as phones. Liquid AI says its 1.7B model beats its larger predecessor while using 35% fewer parameters, its own figures.

Why it matters: Small models that run on one GPU or a phone widen who can build and deploy AI.

Sources: Hugging Face

Acquisitions + deals

Anthropic files for IPO as SpaceX secures $40B chip deal

Anthropic filed its S-1 to go public on Nasdaq under the ticker ANT. The filing shows $16.3 billion in revenue for the first half of 2026 and an $80.7 billion net loss driven mostly by accounting charges. Separately, SpaceX secured roughly $40 billion to lease AI chips through an arrangement with Apollo Global Management.

Why it matters: A record-sized AI listing and a giant chip deal signal how much capital the sector now demands.

Sources: TechCrunch, AnthropicAI, The Information, Reuters, The New York Times, Business Insider

Robotics funding jumps, led by US and China

Crunchbase reports global venture funding for robotics startups reached $9.9 billion in the third quarter, more than double a year earlier. It says China and the US drew most of the money, with Germany a distant third. Large rounds included $1.4 billion for Skild AI and $900 million for China's Xpeng Robotics, Crunchbase says.

Why it matters: Surging robot funding points to where investors expect the next wave of automation.

Sources: Crunchbase News

More funding

• TechCrunch reports Dallas firm Disruptive raised a $10 billion fund to back at least 10 late-stage AI and defense companies. Sources: TechCrunch
• TechCrunch reports October SEC filings show new venture funds, including a $1.252 billion debut AI fund from two former Benchmark partners. Sources: TechCrunch
• GeekWire reports Seattle's Kivo Health raised $10 million in a Series A for its AI-assisted home rehab program for chronic lung disease. Sources: GeekWire
• GeekWire reports Seattle startup Avante raised a $30 million Series A for AI agents that help companies manage employee health benefits. Sources: GeekWire
• GeekWire reports Seattle's Emerald Nuclear won a $2.1 million award from Norway's research council to develop small nuclear reactors for ships and floating power plants. Sources: GeekWire
• TechCrunch reports cloud startup General Compute raised a $100 million Series A, valuing it at $800 million, for a cloud built on inference chips from Nvidia rivals. Sources: TechCrunch
• SiliconANGLE reports Hyphenate raised a $45 million Series A, led by Nexus Venture Partners, for AI agents that automate corporate finance tasks. Sources: SiliconANGLE
• TechCrunch reports Nvidia joined a $14 million round for Indian startup Simplismart, which makes software to run AI models with fewer GPUs. Sources: TechCrunch
• SiliconANGLE reports Bedrock Robotics raised $600 million in a Series C led by Alphabet's CapitalG and Emergence Capital for its autonomous excavator technology. Sources: SiliconANGLE

MCP + agent tooling

GitHub Copilot adds AI-judged command permissions

GitHub made 'assisted permissions' widely available across Copilot's CLI, app and Visual Studio Code. An AI model checks each requested command and automatically approves ones it judges clearly safe, flagging riskier ones for a person. Enterprise administrators can control or turn off the mode through managed settings.

Why it matters: Letting AI clear safe commands can speed coding work while keeping risky actions under review.

Sources: The GitHub Blog

SDKs + APIs

LiteLLM dev build bundles 704 changes

LiteLLM released a development build gathering 704 pull requests. New additions include an admin chat interface with image generation and a feature that continues an interrupted streaming answer on a backup model. It also fixes a bug that had broken tool calls on basic installs since version 1.106.0.

Why it matters: Frequent updates to open tooling change what features developers can rely on day to day.

Sources: LiteLLM on GitHub

Research

OpenAI claims three science results, adds expert reviewers

The Verge reports OpenAI announced three results in math, computer science and physics that it says its internal AI models produced. This time OpenAI asked outside experts to vet the computer science and physics claims. One claimed result disproves the Exponential Time Hypothesis, a widely held assumption in computer science, if the proof holds.

Why it matters: Outside review of AI-generated proofs tests how science absorbs machine-made claims.

Sources: The Verge, The Decoder, Engadget

Study questions AI judges of research novelty

The Allen Institute for AI released a benchmark testing whether large language models can reliably judge if a research idea is novel. Researchers found small wording changes to a prompt could swing judgment accuracy by more than 50 percentage points. The test covered six AI judges, including GPT and Claude models.

Why it matters: Unreliable novelty judging complicates efforts to let AI help propose original research.

Sources: Hugging Face

Other notable news

Apple's 27.2 update expands Siri AI features

Apple released iOS and macOS 27.2, giving Siri a memory feature that recalls past conversations. The update adds Siri AI in five more languages, though it stays unavailable in Europe and mainland China.

Why it matters: New Siri features and languages change what the assistant can do for more users.

Sources: Ars Technica, MacRumors, 9to5Mac

OpenAI case study: Retell powers 100M monthly calls

OpenAI published a case study on voice AI firm Retell, which it says handles more than 100 million calls a month. OpenAI says Retell uses its models for 80% of call responses and Codex across its own development. Retell's own benchmark ranked OpenAI's GPT-6 Astra highest for instruction-following among the models it tested.

Why it matters: Benchmark results from a heavy user offer a real-world gauge of voice-model quality.

Sources: OpenAI

Cloudflare opens PHP hacking challenge for researchers

Cloudflare launched a Capture the Flag challenge inviting researchers to try bypassing its web firewall for two recent PHP flaws. The challenge follows a week in which Cloudflare also responded to compromised certificates and added new firewall rules.

Why it matters: Open challenges on live defenses help test whether firewall protections actually hold.

Sources: Cloudflare Blog

Thanks for reading, and see you in the next edition.

---

About ScrutifAI: ScrutifAI is an independent third-party agency that tests software for AI agents, building a single access point where agents and people can access and use tested tools without learning each one individually, at a fraction of the cost of buying them directly.