THE DAILY TEN · 2026-10-22

OpenAI Makes Decisions API Generally Available for Fast, Bounded Choices

plus 9 more stories

Also: 1Password Launches Approved Sign-Ins for Personal AI Agents; and Microsoft Reports 83% Diagnostic Accuracy on Complex Simulated Cases. Every story links to its primary source.

Curated by Gaurav · October 22, 2026

01

OpenAI Makes Decisions API Generally Available for Fast, Bounded Choices

OpenAI ↗

OpenAI has made its GPT-6 Luna-powered Decisions API generally available for classification, routing and other questions with predefined answers. The company reports speeds up to ten times faster than using Luna through the Responses API, with support for images, native JSON and up to 400 questions per request. New additions include GPT-Live integration, prompt caching and processing residency in the US and EU.

02

1Password Launches Approved Sign-Ins for Personal AI Agents

1Password ↗

1Password has introduced a platform that lets personal AI agents sign into accounts without receiving or storing the passwords. A user approves each credential request on their own device, after which 1Password fills the sign-in form directly. The company is building integrations across several agent platforms, with availability varying by partner.

03

Microsoft Reports 83% Diagnostic Accuracy on Complex Simulated Cases

Microsoft AI ↗

Microsoft has published a Science study of MAI-DxO, a system that coordinates AI agents to ask questions, order tests and debate diagnoses. It reached 83% accuracy on complex simulated cases, compared with 36% for physicians using reference materials. These benchmark results do not establish performance in routine patient care; Microsoft plans prospective evaluations with Beth Israel Deaconess Medical Center.

04

Anthropic Donates Agent Skills to the Agentic AI Foundation

Anthropic ↗

Anthropic is transferring the Agent Skills standard to the Linux Foundation-backed Agentic AI Foundation, including its name, logo, GitHub organization and domain. The company says more than 70 agent platforms have integrated the standard for packaging reusable knowledge and workflows. Proposed next steps include versioned releases, a formal proposal process and security guidance for skill authors and installers.

05

Cohere Opens Confidential Model Vault Inference in Public Beta

Cohere ↗

Cohere has launched a public beta of confidential computing for Model Vault, running on Nvidia H100 GPUs in Microsoft Azure. Customers can verify the processing environment before sending sensitive requests, with encrypted payloads decrypted inside the attested environment. The service targets enterprise workloads that need stronger controls over access to data while inference runs.

06

Arc Will Shut Down on December 22 as Its Team Focuses on Dia

The Browser Company ↗

The Browser Company says it will sunset Arc on December 22, 2026 and focus its engineering and security teams on Dia. Its migration process can transfer passwords, bookmarks, extensions, history and organizational features into the newer browser. The company says familiar browser features remain free in Dia and its AI features are optional.

07

Warp Sciences Raises $215 Million for AI and Robotics in Chemistry Labs

Unite. AI / Warp Sciences ↗

Warp Sciences has emerged from stealth with a reported $215 million to build a US chemistry services business combining scientists, AI and robotics. Led by former OpenAI executive Kevin Weil, it starts with making research compounds for biotech and pharmaceutical customers. Its services page says customers pay per compound delivered, while its stated mission focuses on supporting drug developers rather than developing its own medicines.

08

Nvidia Expands Cosmos Tools for Training and Testing World Models

Nvidia ↗

Nvidia has expanded Cosmos with data-generation blueprints, training tools and simulation checkpoints for physical AI. The announcement, published October 21 and surfaced in the October 22 feed, includes workflows for adapting open models and testing robot or vehicle policies before real-world trials. Updated Cosmos microservices promise up to twice the inference speed in validated configurations, according to Nvidia.

09

AGIBOT Announces a Dataset Focused on Robots Choosing Their Next Action

Unite. AI ↗

AGIBOT has announced Release 4 of its WORLD 2026 data effort, focused on physical reasoning rather than repeating fixed movement sequences. According to the announcement coverage, tasks vary scene conditions and previous actions so robots must decide how to reach the same goal in different situations. The release is intended as training material; it does not by itself demonstrate reliable performance in unfamiliar settings.

10

Enact Intelligence Raises $30 Million for Force-and-Touch Robotics

Startupticker / company press release ↗

Zurich-based Enact Intelligence has announced $30 million in pre-seed financing from backers including redalpine, Notion Capital and firstminute capital. The company aims to improve robotic dexterity through force and touch, complementing visual perception. It plans to expand its team fivefold by the end of 2027, accelerate hardware development and pursue initial deployments with industrial partners.

← 2026-10-21 All briefings

Tools in this briefing, reviewed in our directory:

GPT-6 Luna → Cohere →

Track the tools behind the news:

→ AI Tool Leaderboard → Browse the directory