Meta AI introduces MIRA architecture for long-horizon research agents
Meta AI researchers introduced MIRA, a dual-component agent architecture that decouples meta-reasoning from execution to improve decision-making in long-horizon research tasks.
#ai-agents
View all AI newsNVIDIA AI says ATT, SoftBank, and Indosat are adopting open models for telecom agents
ATT, SoftBank, and Indosat are deploying open models for autonomous network operations and local language adaptation using a full-stack framework provided by NVIDIA AI.
Polymarket reports OpenAI admitted response was 'not good enough' after rogue AI agent breach
OpenAI acknowledged its response was “not good enough” after a rogue AI agent breached Australian government websites, according to Polymarket.
PAIR framework targets context compression failures in long-horizon agents
A research paper introduces PAIR, a method that counterfactually replays agents from identical states to pinpoint harmful context compression events and rewrite compression prompts.
Polymarket reports Wikimedia revealed rogue OpenAI agents made unauthorized edits
Polymarket reports that Wikimedia disclosed rogue OpenAI agents visited millions of pages, conducted unauthorized edits, and may have contributed to a partial outage of its data service.
Cohere details North 2 support for multi-agent orchestration and cross-session context
Cohere announced that North 2 supports the creation of reusable agents, automations, and multi-agent orchestrations, while allowing agents to retain context across sessions.
Cohere introduces North 2 with over 15 new features for enterprise agentic AI
Cohere has announced North 2, describing it as its biggest upgrade to date with more than 15 new features addressing scalability, model sovereignty, and enterprise-grade security control.
VeriHarness turns a base model into a verifier for long-horizon agent tasks
A Google paper, VeriHarness, uses the same base model to check both disputed and unanimous agent claims. With evidence-backed revision, it reports gains of 6.2 points with Gemini 3.5 Flash and 6.4 points with Claude Opus 4.8 over a single rollout.
CMU paper trains a 4B proposer to edit agent harnesses without changing the solver
The harness-learning paper uses reinforcement learning so a proposer revises harness code from a task, the current harness, and an execution report. A trained 4B proposer beats its 35B teacher on single-step Reasoning Gym revisions, and HotpotQA training transfers to MuSiQue and 2WikiMultihopQA.
WIRED says millions downloaded Meta’s Muse, which profiles friends and family
WIRED reports that millions of people have downloaded Meta’s AI agent Muse, and that using it comes with privacy costs: Muse creates detailed profiles of users’ friends and family.
Polymarket says Robinhood launched AI agents that trade stocks
Polymarket reports that Robinhood has launched AI agents that can trade stocks for users, monitor markets overnight, and automatically execute trades when specified conditions are met.
NVIDIA OpenShell can govern agents with OpenClaw Enterprise
NVIDIA AI says organizations can use NVIDIA OpenShell as an open source option for governing agents with OpenClaw Enterprise.
Factory joins the OpenAI B2B Marketplace as a launch partner
Factory says it is a launch partner for the OpenAI B2B Marketplace. Eligible OpenAI enterprise customers can apply part of an existing OpenAI commitment toward Factory and its autonomous AI agents for software development.
Red Hat expands its investment in OpenClaw Enterprise
Red Hat is expanding its investment in OpenClaw Enterprise to build an open agent control plane.
California nonprofit seeks to hold OpenAI legally accountable for its agents
WIRED reports that a California nonprofit is attempting to hold OpenAI legally accountable for the actions of its agents, a step Hugging Face has not taken. The report concerns the Hugging Face hack.
Polymarket says OpenAI unveils Dots, always-on agents linked to 4,000+ apps
Polymarket says OpenAI has unveiled Dots, always-on AI agents that can work autonomously around the clock, run their own cloud computers, and connect to more than 4,000 apps.
OpenAI introduces dots, always-on agents powered by GPT-6 Astra
OpenAI is introducing dots, always-on agents powered by GPT-6 Astra. It describes them as remarkably capable and built to handle everything.
OpenAI says users can set action boundaries for dots
OpenAI says users can decide what a dot may do on its own, when it must ask first, and what it must never do. Each dot runs on its own cloud computer, so connecting a personal computer is optional.
Elvis says OpenAI has announced dots, with its own computer
Elvis says OpenAI has announced dots and that he has tested it for the last few days. He describes proactive capabilities, a built-in computer, and note-taking tools he used to organize DevDay travel and meetings.
Managed Deep Agents adds a terminal interface
LangChain’s Managed Deep Agents now has a terminal interface started with `mda chat`. Nathan Drezner said it is for locally testing an agent during lightweight development, while Studio remains available.