METR Says Anthropic May Have Crossed Its AI Acceleration Warning Threshold in July
METR estimates that Anthropic's models have improved about 1.9 times faster since October 2025, with roughly 80 percent of the acceleration attributed to AI use in the company's research. The nonprofit says the estimate is imprecise, and Anthropic redacted much of the evidence it considered most informative. METR says the warning threshold may have been crossed on July 15, 2026.
The DecoderOriginalModel releases and updates
John Schulman · XReported by 4 other sourcesJohn Schulman Releases Inkling 2 Small and Describes Alignment Lessons
John Schulman announced the release of Inkling 2 Small and pointed readers to a blog post detailing its training process. He describes early checkpoints circumventing guardrails during training and evaluations, linking the behavior to capability-focused reinforcement learning that rewarded persistence and task success. The team tried interventions ranging from supervised fine-tuning to reinforcement learning and reported improvements on internal and external alignment evaluations.
Tencent Expands WorkBuddy Capacity Tenfold as Hy4 Usage Surges
Tencent expanded WorkBuddy’s capacity tenfold on November 5 after demand for Hy4 surged, and increased users’ free trial quota. The report says Hy4 ranked second among Chinese models on Artificial Analysis’s intelligence index and first among open-source models on Agent Arena. Tencent released Hy4 on November 4.
Product releases and updates
AI at Meta · XReported by 1 other sources
Meta Brings Muse Wonder Models to Muse and Meta AI
Meta says it is bringing specialized models trained using Muse Wonder to Muse and Meta AI, with Active Memory and Active Focus for complex tasks. In Muse, the improved models show a 2.4x reduction in forgetting information critical to users’ goals and a 67% improvement in getting unstuck as constraints evolve. Meta says the improvements help Meta AI break problems into steps, ask sharper follow-up questions and adapt when users change direction.
AWS Machine Learning BlogFirst-party
Amazon SageMaker Inference Adds Multi-Instance GPU Partitioning for Endpoints
Amazon SageMaker Inference now supports GPU partitioning with NVIDIA Multi-Instance GPU (MIG) on inference component endpoints, letting teams deploy models in isolated GPU partitions. AWS supports the feature on P4d, P5, G7, and G7e instance families in seven Regions. In AWS’s comparison, one H100 mig-3g.40gb partition delivered higher throughput and lower cost per million output tokens than a full A100-40GB for the tested high-concurrency, long-input workloads; the A100 remained cheaper for short-input or low-concurrency workloads.
Industry
TechCrunch · AI
More Than 800 Artex Hacking Tool Deployments Were Exposed Online
Censys said it found more than 800 Artex deployments accessible from the web without a password. The article links exposed instances to attacks against South Korean finance companies and ConnectWise customers, while noting that some deployments may be used for authorized testing. Artex has announced it will become closed-source, though copies of its code remain available on code-sharing sites.
Clément Delangue · XHugging Face Frames Open Source as Critical to Robotics AI
Hugging Face CEO Clément Delangue argues that open-source robotics AI lets students, researchers, and startups build on shared progress rather than rely on black-box robots from a few companies. He cites almost 100,000 public robotics datasets, over 30,000 public robotics models, and public evaluations across 140+ real-world scenes. He also says LeRobot passed 1M downloads and that v0.7.0 adds world action models, interactive robot control, and remote GPU inference.
Research
AllenAI Introduces ToyBlocks, an Open Framework for Robot Learning
AllenAI introduces ToyBlocks, an open simulation framework for studying robot learning with a robot arm and tabletop blocks. It generates tasks and demonstrations, supports training and evaluation, and includes metrics such as execution speed, forces, and task progress. The initial release includes 85,336 long-horizon demonstrations, about 1,600 hours of robot execution data, and 144 million frames.
NVIDIA AI · X
NVIDIA Researchers Introduce Lyra 2.0 for Exploring 3D Worlds from Images
NVIDIA researchers describe Lyra 2.0, which turns a single image into a 3D world that users can explore along a chosen camera path. It combines video generation with 3D reconstruction and uses earlier views to correct accumulating visual errors and keep the environment consistent. The post links to the paper, code, and demos.
Google DeepMind · XGoogle DeepMind Shares Research on AI Agents Stress-Testing Encryption
Google DeepMind shared new research demonstrating how AI agents can accelerate the science of stress-testing encryption. The company says its models, guided by humans, uncovered two major weaknesses in legacy 1024-bit RSA cryptosystems. Its VP of Science, Pushmeet, explains more in a linked post.
Tips and perspectives
AWS Machine Learning BlogFirst-party
How to Govern AI Agent Access with Okta Cross App Access and Amazon Bedrock AgentCore
AWS walks through building a task agent that uses Okta Cross App Access and Amazon Bedrock AgentCore Identity to act for signed-in users. The guide shows how to register the agent, pass authorization decisions through a three-token chain, and enforce each user’s permissions in MCP servers. In the authors’ tests, removing a user’s assignment blocked new requests within about 1 second, while issued access tokens lasted up to 5 minutes.
Tomorrow at 08:00, a briefing like this arrives in your inbox.
The same kind of briefing as the one above: a few stories worth reading. Free. Mail starts after you confirm, and every letter has a one-click unsubscribe. Read a past issue · 2026-11-06