UK Unveils AI Biosecurity Lab; Anthropic Ships Proactive Agent
The AI software race sees a new focus on safety collaboration and proactive agents, while key players expand cloud reach and face investor scrutiny over spending.
The story
The UK government has launched a new National AI Biosecurity Lab (AIBL) in a significant move towards AI safety, backed by major labs. OpenAI and Anthropic are each contributing up to £100 million in model credits and access, while Google DeepMind will collaborate on research with Basecamp Research.
This initiative underscores growing concerns about the safe development of advanced AI. Concurrently, Anthropic introduced Claude Duet, a proactive agent that monitors emails and calendars, handles tasks, and seeks approval for sensitive actions, marking an evolution in agentic capabilities.
Meanwhile, OpenAI is expanding the availability of its models, including GPT-6 Astra and GPT-6.1 Sol, to Google Cloud customers through a private preview, diversifying its cloud partnerships beyond Microsoft and Amazon. On the safety front, OpenAI also implemented an internal restriction, cutting off its own frontier training from live internet access. These developments reflect a dual emphasis on pushing AI capabilities while establishing guardrails and broader accessibility.
Who moved
Anthropic
What Changed: The company launched Claude Duet, a proactive AI agent operating on its Opus 5.5 model, designed to manage tasks and provide daily briefings based on connected user data.
Consequence: This marks a step towards more autonomous and integrated AI assistants, requiring user approval for sensitive actions like payments.
OpenAI
What Changed: OpenAI announced its models, including GPT-6 Astra and GPT-6.1 Sol, will be available on Google Cloud, starting with a private preview.
Consequence: This expands OpenAI's cloud ecosystem beyond its primary partnership with Microsoft and previous deal with Amazon, making its models accessible across all three major cloud platforms.
DeepSeek
What Changed: The Chinese AI lab is reportedly raising approximately $15 billion in funding.
Consequence: This significant capital injection comes from major Chinese industrial players such as CATL, Tencent, Meituan, and Geely, signaling strong investor confidence in the region's AI development.
Moonshot AI
What Changed: Moonshot AI introduced Kimi K3.1 Pro, a post-trained upgrade in its K3.1 series, featuring 2.8 trillion parameters and a 1-million-token context.
Consequence: The model achieved a 65.7% score on Terminal-Bench 4.0 and is designed for complex coding, 3D modeling, and finance applications, with open weights expected by November 4.
Meta
What Changed: Meta reported its third-quarter revenue increased by 26%, but its shares declined due to an increased AI spending forecast for 2026 and plans for new stock sales.
Consequence: The company's projected 2026 capital expenditure rose to $135 billion to $145 billion, resulting in a $2.12 billion free cash flow deficit for the quarter.
New models
Claude Duet
Lab: Anthropic
What: A proactive AI agent built on Opus 5.5 that monitors digital workspaces and handles tasks.
Use: It is intended for Pro and Max subscribers to automate scheduling, generate briefings, and perform other background tasks with user approval.
Kimi K3.1 Pro
Lab: Moonshot AI
What: A 2.8 trillion-parameter model with a 1-million-token context window, achieving a 65.7% score on Terminal-Bench 4.0.
Use: Designed for long coding tasks, 3D modeling, and finance applications, with open weights planned for release on November 4.
M3.1 Flash (Open Weights)
Lab: MiniMax
What: The open-weight release of a 428-billion-parameter multimodal Mixture-of-Experts (MoE) model.
Use: Developers can download and deploy this model for coding and AI agent tasks, benefiting from its text, image, and video processing capabilities.
Market signals
DeepSeek is raising approximately $15 billion from major Chinese companies including CATL, Tencent, Meituan, and Geely.
Implication: This significant funding round highlights continued investor confidence and substantial capital allocation towards AI development in China.
Meta reported a $2.12 billion free cash flow deficit in Q3, following an increase in its 2026 AI capital expenditure forecast to between $135 billion and $145 billion.
Implication: The substantial investment in AI infrastructure is impacting Meta's short-term financial liquidity, indicating the high cost of competing in the AI race.
Modal Labs secured $750 million in Series D funding, valuing the company at $15.75 billion.
Implication: This large raise for an AI infrastructure platform demonstrates strong investor appetite for companies building the foundational compute and tooling layers for AI applications.
Flapping Airplanes, an AI lab focused on low-cost model training and operation, closed a $1 billion Series A funding round.
Implication: The substantial early-stage funding suggests investor interest in efficiency and cost reduction within foundational AI model development.
What we'll be watching
- Moonshot AI is scheduled to release the full open weights for its Kimi K3.1 Pro model on November 4.
- Google's introductory pricing for Gemini 4 Argon is set to double on December 1, moving from $2/$10 to $4/$20 per million input/output tokens.
- Discussions continue among OpenAI, Google, and Anthropic regarding the establishment of a 'Standards Authority for Frontier AI' by the end of this year or early next year.
Reporting + analyst voices: grounded via Google Search at publish time.