join.dev
AI news · 10 stories
OpenAI launches ChatGPT Wallet for user-approved purchases · LangChain raises $300 million to expand agent infrastructure · Salesforce releases SOV-Transcribe-1 for controllable speech recognition
OpenAI launched ChatGPT Wallet on November 13 for eligible adult US Pro subscribers using dots. Each purchase requires user review and a security check, while payment tokens keep actual card details away from agents and merchants.
Read at OpenAI
LangChain announced a $300 million Series C at a $3 billion valuation on November 13, led by CapitalG. The funding supports its work on infrastructure for running, improving and governing enterprise agents through LangSmith.
Read at LangChain
Salesforce released SOV-Transcribe-1 on November 12 under Apache 2.0 for non-streaming speech recognition. Its instruction-controlled formatting and contextual keyword support give developers more control over transcription on their own infrastructure.
Read at Hugging Face
Google Research reported November 13 results from Antigravity’s /teamwork collaborations with human experts across mathematics, chemistry and systems engineering. The work includes formal verification and kernel optimization, illustrating how coordinated agents can support research and infrastructure development.
Read at Google Research
NVIDIA’s November 13 walkthrough combines reusable SimReady robot and object assets in Isaac Sim for a beverage-can pick-and-place task with agent-generated controls. Authored physics properties and validation evidence help developers compose and check simulation scenes.
Read at Hugging FaceGitHub made Enterprise Server support generally available in Copilot app 1.2.0 and later on November 13, bringing remote repositories, issues and pull requests into the app. Users must first sign in with a GitHub.com or GHE.com account; cloud sessions remain unavailable for Enterprise Server.
Read at GitHub
GitHub announced on November 13 that selected Copilot models will retire on December 14 and published suggested replacements. Teams should update integrations and model policies; Claude Haiku 4.5 remains available to individual subscribers on annual plans.
Read at GitHub
Anthropic’s November 12 report says its Anvil system found stronger jailbreaks more efficiently than expert red-teamers in internal safeguard evaluations. The system relies on privileged model access and testing conditions, so the findings demonstrate a defensive evaluation method rather than equivalent capabilities for ordinary users.
Read at Anthropic
Anthropic’s November 12 release notes introduced self-hosted Claude Managed Agents, with access requested through account teams. Customers can keep the agent service and session history on their infrastructure while sending model requests to a chosen provider, separating service hosting from inference.
Read at Anthropic
Vercel announced Workflow SDK 5 on November 12 with encrypted payloads, regional run placement and lower execution overhead for automation and agent loops. Configured failover can move step execution to a fallback region, while run state stays in its original region.
Read at Vercel