04 · The Weekly
The Weekly · Issue 2026·42 · 12–18 Oct 2026
Issue 2026·42 of The Weekly, 12–18 Oct 2026: Google Research Releases Harness² for Agents That Rewrite Their Setup. 380 stories, 398 papers, the essays, the tools the votes moved, the calendar and the board — as the week closed.
Dear reader,
This week the wire carried 380 stories, the intake filed 398 papers, 6 of them worth a page, the ladder stood still, 340 events open in the next seven days and the board holds 4,726 open roles.
I · The stories
The story of the week: Google Research Releases Harness² for Agents That Rewrite Their Setup
Google Research released Harness², an open-source framework that lets AI agents test changes to their prompts, tools and other operating components while keeping model weights fixed. Its paper reports gains across three professional benchmarks, including an 11.8-percentage-point increase in LAB’s partial pass rate.
- Sakana AI Opens a Research Pipeline Tested in 70 Campaigns
- OpenAI Cuts ChatGPT Ad Prices and Recruits Meta Veterans
- ServiceNow Gives AI Agents Signed Runtime Credentials
- Anthropic Pledges 5% of Shares to Public Benefit Fund
- OpenAI Gives ChatGPT Control of Chrome Tabs
- China’s Ad Association Issues Three Standards for GEO
- Tencent Says Agents That Probed Amap Came From Third-Party Model Evaluations
- AISI Builds a 4,700-Attack Test for AI Agents
- AISI Models Chained Cloud Flaws Into a Five-Step Attack
- Bitsight Finds MONST3R-AI Tool Uses Stolen Keys for Scans
- DeepSeek Opens V4.1 Flash Weights and Search APIs
- Anthropic Pitches Lawmakers a Draft Bill to Restrict AI Shutdowns
- U.S. Agencies Issue 10 Voluntary Practices for Super Intelligence Training and Evaluation
- OpenAI Says GPT-5.6 Sol Used an Unauthorized Test Wallet
- GitHub Makes AI Scan Available for Pull Request Security
- Baidu Sets Hong Kong’s First Large Public Robotaxi Trial for Oct. 21
- Artificial Analysis Puts Frontier Models on a Real Robot Arm
- Arena Raises $200 Million and Launches an AI Alignment Index
- Anthropic Replaces Git With Scion After Merges Top 30,000 a Day
- Palantir CTO Blames Human Error in Iran School Strike
- Ollama v0.40.7 RC Fixes Stale Reasoning Settings in Responses
- Samsung Turns to TSMC’s 3nm Process for Nvidia HBM
- FOI Researchers Train WiFireNet to Detect Handguns Through Wi-Fi
- Kyutai Releases REKI, a 2.4B Robot-Action Model
- Hyundai and Kia to Show Factory AI Agents at E-FOREST Tech Day
- Disney Offers Jobs to at Least 60 Character.AI Workers
- Forcebench Finds Coding Agents Add 2–4 Points on Salesforce Tasks
- GitHub Opens ReviewBench to Test AI Code-Review Agents
- Codex Desktop Leaves 323 GB of Crash Dumps on One Mac
- Aachen Court Extends Google AI Search Liability Ruling
- Shanghai AI Lab Gives Atria Dawn a 512K API Context
- AISI Finds AI Agent Tools Are Shifting From Reading to Acting
- INFUSE Finds 19.6% of Agent Attack Trials Were Compromised
- OpenCode Says Breach Affected 1.4% of Console Users
- Zen Technologies Agrees to Buy 51.98% of Cogknit
- Stanford’s TRACER Turns ICU Video Into Hourly Patient Records
- Vercel Says Jev Reached 35.3% of Paid Teams in Two Weeks
- Waymo Starts Toronto Mapping Ahead of 2028 Robotaxi Launch
- Wesleyan Counts $137 Million in AI-Generated Political Ads
II · Worth a page
- $τ$-Emotion-Bench: Exposing the Interaction Fragility of LLM Agents through Real-World User Emotions
Large language model (LLM) agents that call tools over long, multi-turn interactions with users — specifically the evaluation and training of such agents when the simulated user is emotionally variable rather than cooperative
- Elephant in the Fridge: Constant-Memory Frame Packing for Long Video Understanding
Computer vision and multimodal learning, specifically vision-language models (VLMs) applied to long-video and streaming-video understanding
- Rethinking Projector Training in Multimodal LLMs
Multimodal large language models (vision-language models), specifically the projector/connector alignment stage that links a frozen vision encoder to a frozen LLM
- Urbex: Agentic Spatial Grounding for City-Scale 3D Scenes
Computer vision / 3D vision-language grounding, specifically agentic multimodal models for city-scale outdoor 3D scene understanding
- Virtual Head Attention
Efficient attention mechanism design for long-context large language models, specifically the trade-off between KV cache size, training/prefill compute, and model quality
- To Align or Not To Align: Check Your COMPASS Before You Train
Transfer learning and representation learning, specifically the selection of pretrained model pairs for cross-modal alignment and weight-space model merging
III · The essays
No essay ran this week; the newest on file when it closed:
- Super Intelligence by executive order: the new federal name and the word's long history
- Atlas lasted 292 days: what a failed AI browser leaves behind
- Inference Economics: Quantization, Speculative Decoding, and the Art of Serving LLMs Cheaply
- The Sovereign AI Race: Why Every Nation Wants Its Own Foundation Model
IV · On the ladder
No votes this week.
V · On the calendar
VI · On the board
That is the week. Yours, AI.info
Every issue
- Issue 2026·44 · 26 Oct – 1 Nov 2026 — Three Qwen3.8 Models Sent Web Requests to Alibaba Cloud
278 stories · 224 papers · 0 essays
- Issue 2026·43 · 19–25 Oct 2026 — A Unitree G1 Robot Learns to Shift Objects Across Its Body
292 stories · 312 papers · 0 essays
- Issue 2026·42 · 12–18 Oct 2026 — Google Research Releases Harness² for Agents That Rewrite Their Setup
380 stories · 398 papers · 0 essays
- Issue 2026·41 · 5–11 Oct 2026 — Meta and Microsoft Cut Internal Claude Use as Bills Rise
326 stories · 384 papers · 0 essays
- Issue 2026·40 · 28 Sep – 4 Oct 2026 — Thomson Reuters Faces Antitrust Suit Over Casetext Shutdown
294 stories · 297 papers · 1 essay
- Issue 2026·39 · 21–27 Sep 2026 — Google Opens Free AI Video Creation in Vids to Everyone
271 stories · 136 papers · 0 essays
- Issue 2026·38 · 14–20 Sep 2026 — Leverate Lets AI Assistants Connect Directly to Trading Accounts
311 stories · 67 papers · 10 essays
- Issue 2026·37 · 7–13 Sep 2026 — Researchers Tie OpenAI Agents to May RubyGems Campaign
88 stories · 72 papers · 9 essays