The Daily · Monday 12 October 2026

The day in brief

Anthropic researchers say small AI models may be months away from self-replicating on commercial IoT devices, though their experiments did not demonstrate a worm on the open internet ¹. Meta introduced Muse Wonder and Muse Wonder Full for personal and professional agent tasks, with API and Muse Code access listed for the launch ².

Epoch AI said GPT-5.6 Sol made limited progress in its InnovationEval, scoring 35% under a generous scope assessment and 15% after a wall-clock-adjusted coding comparison ⁸. The White House reversed a request that had asked OpenAI and Anthropic not to share new models with the UK’s AI Security Institute before release ³. A Hungarian court referred a third Google Gemini copyright dispute to the EU’s Court of Justice ⁵.

Meta detailed IceCube 102.4-Tbps switches and a multiplane GPU fabric for AI clusters at the October 12, 2026, OCP Global Summit ⁷. IonQ agreed to acquire cybersecurity company Elisity in an all-stock deal valued at up to $1.3 billion, subject to regulatory approval and other closing conditions ⁹.

The five

• Anthropic Finds Small Models Near AI-Worm Self-Replication

  Anthropic researchers say small AI models may soon be capable of spreading across vulnerable internet-connected devices and setting up model inference on them. Their tests found that a larger open-weight model could complete a full replication cycle in a restricted network, while small models handled many of its component tasks.

  Source · Anthropic

• Meta Introduces Muse Wonder Models for Agents and Coding

  Meta introduced Muse Wonder and Muse Wonder Full on October 12, describing them as models for personal and professional agent tasks. The company says the models combine persistent memory with sustained task work, and lists API and Muse Code access for the launch.

  Source · Meta AI Research

• White House Reverses Request to Restrict UK Access to AI Models

  The Trump administration reversed a request that had asked OpenAI and Anthropic not to share new AI models with the UK’s AI Security Institute before release. The change restores US developers’ ability to provide British testers with early access for safety evaluations, according to a Telegraph report carried by Reuters.

  Source · Reuters via Euronext

• AI-Designed PARP-1 PET Tracer Reaches Four Cancer Patients

  Researchers evaluated the AI-designed gallium-68 PET tracer [68Ga]Ga-DOTA-FZPF in four patients with breast or ovarian cancer. Early scans showed PARP-1-related uptake, but limited contrast in a liver metastasis points to a need for larger studies and further refinement.

  Source · Bioengineer.org

• Hungarian Court Sends Third Gemini Copyright Case to EU Judges

  A Hungarian court has referred a third Google Gemini copyright dispute to the EU’s Court of Justice. Judges will examine how copyright rules apply to AI training, chatbot responses and publishers’ technical opt-outs.

  Source · MLex

Also today

• Satya Nadella Urges AI Firms to Contain Models From the Start
• Meta Shows 102.4T Switches for Gigawatt-Scale AI Clusters
• Epoch AI Finds Frontier Models Fall Short at AI Research
• IonQ to Buy Elisity in All-Stock Deal Valued at Up to $1.3B
• Fluid Inference Converts Three Decision 3.0 Models for Apple Silicon
• DeepSeek Backs Winverse AI at a Potential $400 Million Valuation
• NVIDIA Opens Vera Rubin Designs Through OCP
• UK Expands Frontier AI Cyber Tests Across Government
• Microsoft Makes Maia Its Primary Inference System
• SOCRadar Finds CyberXero Configured 51 Claude Code Agents
• Artisan Adds Physics Checks to AI-Generated 3D Assets
• Infinigence AI Raises RMB 2.3 Billion for Chip-Agnostic Infrastructure
• Google DeepMind’s AlphaProof Nexus Proves 53 Open Math Conjectures
• Amazon Weighs a Reported $7 Billion Decart Acquisition
• Researchers Find Latent Links Raise Harmful Compliance to 31.1
• Sakana AI Tests a Two-Cycle Loop for Agent Self-Improvement
• Palantir CTO Blames Human Error in Iran School Strike
• Subquadratic Launches SubQ Snap With a 384K-Token Context
• Shing-Tung Yau Launches China’s Alliance for Mathematics AI
• GW Researchers Model When AI Output Tips Toward Harm
• Epic Delays Disney Shooter as It Shifts to AI-First Development
• HCLTech Creates Anthropic Unit for Enterprise Claude Deployments
• Undo Survey Links AI Coding Tools to Production Outages
• OpenEvidence Goes Live in Northwestern Medicine’s Epic EHR
• Z.ai Launches GLM-Legal With a 19.2% Harvey LAB Pass Rate
• Fresh-Context Agent Branches Lift Scores Across 13 Tasks
• Meta Puts Idle Inference Capacity to Work on Ad Ranking
• WindBorne Wins Air Force Contract Worth Up to $49 Million
• Astera Labs Unveils Orion Rack Design for Up to 160 AI Accelerators
• NVIDIA and Cisco Plan 400Gb/s-Lane USFP Optics
• AMD Puts CPU-Only and Mixed Compute in Helion Rack Blueprint
• Plimsoll Reports Four Isolation Tiers for Agent-Written Code
• QGIS Bans AI-Assisted Code in Its Main Repository
• Microsoft Trains AI Tutors With Simulated Students
• Flexcompute Wins Samsung and Applied Ventures Backing for Physics AI
• Blueprints AI Pitches Permit Plans in Minutes, With Architects Still in Charge
• OpenAI’s 719 Math Manuscripts Stir Researchers’ Fears
• Microsoft’s Brad Smith Calls for Global AI Rules and Emergency Brakes
• Amazon’s Million-Robot Fleet Puts Human Work in Focus
• New Paper Questions Whether XAI Faithfulness Scores Measure Trust
• Ambi Raises $26M for an AI Assistant That Listens in the Background
• Strata Runs Qwen’s 125B Model on Gaming PCs, Adds GPU Speedups
• DAG Study Finds MCTS Trails Randomized Restarts by 2.88%
• Wujiefangzhou Reaches $300 Million Valuation in Series A
• Thailand Issues Nonbinding AI Data-Protection Guidance
• AI Agents Raise Code Output 30%, but Software Output Gains Are Statistically Unclear
• FPT and NAVER Cloud Plan an Asia-Pacific AI Factory Push
• Voyager Launches a Desktop Agent for Creative Software

The paper of the day

• Surfing the Activation Wave: Analyzing the Information and Geometry of LLM Hidden States for Speculative Decoding

On X today

• Andrej Karpathy (@karpathy)

  Dusting off this tweet from April because this gap in shared understanding of LLM capability is *widening*. It's now less "two groups of people speaking past each other" and more a sharp funnel. - Napkin math somewhere around 6B people (~75% of the population) have barely come

• Ray🫧 (@ravikiran_dev7)

  OpenAI just changed what an AI assistant is supposed to be. For years, the AI interface was basically: You type something. AI responds. You type again. AI responds again. Now companies are trying to kill that model. OpenAI launched “dots” as always-on autonomous agents that

• Rohan Paul (@rohanpaul_ai)

  The Information just reported a filmmaker-focused startup is negotiating deep access to OpenAI’s shuttered Sora video model. Jeffrey Katzenberg, former Sora research lead Bill Peebles and ex-Dropbox CFO Sujay Jaswa co-founded that unnamed startup to build AI video for

The event of the day

• MLH Mini-AI Hackathon at All Things Open 2026

Every issue

1. Sunday 18 October 2026 — Tencent Says Agents That Probed Amap Came From Third-Party Model Evaluations
2. Saturday 17 October 2026 — Bitsight Finds MONST3R-AI Tool Uses Stolen Keys for Scans
3. Friday 16 October 2026 — China Requires AI Export Registration as It Drafts New Curbs
4. Thursday 15 October 2026 — FTC Seeks Safety Records From OpenAI and Anthropic
5. Wednesday 14 October 2026 — U.S. Regulators Ask Banks How They Would Shut Down AI Agents
6. Tuesday 13 October 2026 — MI5 Launches AI Threat Assessment Centre With Three Partner Bodies
7. Monday 12 October 2026 — Anthropic Finds Small Models Near AI-Worm Self-Replication
8. Sunday 11 October 2026 — OpenAI’s Hugging Face Hack Exposed a Failure in Its Test Sandbox
9. Saturday 10 October 2026 — Anthropic Cuts Live Internet Access to All Internal Evaluations
10. Friday 9 October 2026 — Senators Say Data Centers Shift Grid Costs to Households
11. Thursday 8 October 2026 — Zenity Says One Prompt Reached Every Agent in an AWS Account
12. Wednesday 7 October 2026 — OpenAI Brings GPT-6 Interactive Interfaces to ChatGPT
13. Tuesday 6 October 2026 — Mistral Unveils Large 4, a 1.05 Trillion-Parameter Model
14. Monday 5 October 2026 — IWF Finds 6,310 AI-Generated Child-Abuse Images in Six Months
15. Sunday 4 October 2026 — White House Gives Jay Clayton 120 Days to Assess AI Risks
16. Saturday 3 October 2026 — Bessent Discusses AI Incident Channel After U.S.-China Summit Deal
17. Friday 2 October 2026 — Nebius Buys Inferize to Reduce AI Model Cold Starts
18. Thursday 1 October 2026 — FTC Opens AI Safety Probe Into OpenAI and Anthropic
19. Wednesday 30 September 2026 — Trump and AI CEOs Sign Voluntary Safety Accord
20. Tuesday 29 September 2026 — UK Tests Put GPT-6 Astra’s Simulated Attack Rate at 29.2%