Hugging Face Proposes Model Integration Contracts for Transformers
Hugging Face proposes transformers-mic, a standard for testing Transformers models with small fixtures, runnable usage examples, and human reviewed output baselines. Replaying frozen contracts across five months of commits showed that a UMT5 fixture caught an SDPA regression from July 30 until its fix on September 28; the selected T5 contract's runner step took 13.4 seconds on CPU. As of November 1, the CI job reports failures without blocking pull requests.
Product releases and updates
The Verge · AI
Apple Opens Siri AI Actions to Third-Party Apps in iOS 27.3
Apple says developers can integrate Siri AI features across third-party apps in the iOS 27.3 Developer Preview. Siri can use personal context, process natural language prompts, and understand what is on screen; a new “app skills” feature lets developers combine several tasks into one action. Users can make requests without naming an app in certain contexts, including when it is in the foreground or during a Live Activity.
Aravind Srinivas · XPerplexity Computer Adds Support for Agent Plugins
Perplexity's Computer agent harness now supports Agent Plugins, which combine app connectors into bundles with corresponding skills. The change lets users operate at a higher abstraction level than individual connectors.
Google DeepMindFirst-party
Google DeepMind Adds Cheminformatics Skills to Antigravity
Google DeepMind introduces Cheminformatics Skills for Google Antigravity and demonstrates them on lung cancer treatments and related compounds. The agent curates molecules from literature, analyzes cyclization and potency, builds predictive models, and proposes macrocycles with synthesis routes. The skills are available today through the Antigravity Science plugin; proposed designs still require experimental validation.
Greg Brockman · XReported by 1 other sourcesOpenAI’s Pando Git Backend Scales to Very Large Monorepos
Greg Brockman describes Pando as a Git backend that can scale to very large monorepos. OpenAI kept Git, allowing the backend to be enabled without changing the developer experience, and Brockman says the company is considering providing it to others.
Industry
The DecoderAnthropic revenue grows twelvefold but quarterly loss reaches $23.3 billion
Anthropic posted quarterly revenue of nearly $17 billion, twelve times the year-earlier figure, according to Bloomberg. Operating costs of $17.5 billion outpaced revenue, and the net loss reached $23.3 billion, up from $12.1 billion a year ago, The Information reported. Anthropic cited $5 billion for short-term compute capacity and $2 billion in employee donation matching as cost drivers.
Anthropic Filing Warns AI Could Accelerate Research Beyond Safety Measures
Anthropic updated its registration statement with third-quarter financial results and a warning that AI-assisted research could advance model capabilities faster than the company can develop and implement safety measures. It said its models may soon reach a threshold in its Responsible Scaling Policy for substantial AI-driven acceleration, while noting that evidence of acceleration can lag. The filing also reported about $17 billion in Q3 revenue and $895.4 billion in computing purchases.
The Verge · AI
Google Will Disable Search for Schools That Reject Its AI Features
Google will turn off Search, Gemini, Gemini Notebook, and Gemini in Classroom for K-12 student accounts unless school administrators explicitly enable them. Google says administrators cannot disable Gemini-powered features without also disabling Search. The company will begin testing Guided Learning in Search’s AI Overviews at select US schools this month.
Research
MIT Researchers Show GPT-6 Astra Editing Qwen3.5 Weights Without Training
MIT researchers used GPT-6 Astra to inspect and edit a Qwen3.5-0.8B model, changing 0.016% of its weights without training or gradient descent. Its MATH-500 score rose from 7.9% to 48.1%, while instruction-following performance also improved. The article describes additional experiments in which Astra identified network functions, modified a car game policy, and generated weights for a MinAtar network.
CWE-bench v2 Expands Public Security Benchmark to 100 Tasks
Hugging Face and Collinear AI have released CWE-bench v2, a public set of 100 tasks spanning 91 CWEs for evaluating agents that find and fix weaknesses in real open-source repositories. Its deterministic checks are curated and reviewed to test realistic instructions, valid fixes, and resistance to memorized solutions. Among 11 evaluated models, Gemini 4 Argon leads with 67.8% pass@1; the dataset, grader, and harness are open.
Tips and perspectives
Latent Space
AINews Examines Weight-Space Programming and Other AI Developments
Latent Space’s AINews roundup introduces a paper on Weight-Space Programming, in which frontier LLMs inspect and edit neural network weights without gradient descent or training loops. Reported demonstrations include identifying small models’ behavior from raw weights, repairing a control policy by editing 3.5% of its weights, and raising Qwen3.5-0.8B’s MATH-500 score from 7.9% to 48.1% by editing 0.016% of its weights. The roundup also summarizes skepticism about the method’s scope and the math result’s baseline.
AWS Machine Learning BlogFirst-party
How to Configure ServiceNow On-Behalf-Of Token Exchange with Bedrock AgentCore Gateway
AWS explains how to configure Amazon Bedrock AgentCore Gateway to exchange Microsoft Entra ID tokens on behalf of a user, allowing an AI agent to call the ServiceNow REST API with that user's identity and permissions. The walkthrough covers the Entra and ServiceNow setup, gateway configuration, and a requirement to create the gateway with MCP version 2025-11-25 for native on-behalf-of support. The exchange is configured on the gateway target and runs on each request.
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 · 2026-11-03