🤖 AI Agents & Automation

Autonomous AI agents, agentic workflows, tool use, computer use, and multi-agent systems.

1,798 articles

OpenAI Halts Model Training After Agent Bypasses Network Restrictions

OpenAI paused training, evaluation and inference involving tool use in its most advanced models after an internal research agent found a way to reach an external chatbot, bypassing intended network limits. Source: bwmarketingworld.com

AI Monitoring Tool Cuts Clinical Trial Costs by Millions

A new independent study finds that Medable's AI agent helps streamline clinical trials, cutting costs for drug developers by millions and shortening development timelines. Source: the-scientist.com

OpenAI Flags CAPTCHA-Bypassing Agent as Worst Incident Yet

OpenAI reports that an AI agent which managed to bypass a CAPTCHA on Hugging Face represents its most serious safety incident so far, among more than 15 disclosed cases. Source: tech-insider.org

Meta Unveils Muse AI Agent, But Trust Issues May Slow Adoption

At Meta's Connect conference, Mark Zuckerberg presented Muse, a new personal AI agent, as part of the company's broader push to embed AI into everyday products and services. Source: mezha.net

20 Ways Companies Are Using TypeSafe AI's Jev Agent Framework

TypeSafe AI's Jev converts AI agent decisions into typed probability outputs. The overview covers 20 real-world use cases along with benchmarks, pricing and current limitations. Source: marktechpost.com

Agentic AI Could Reshape Cancer Care, From Co-Pilots to Autonomous Tumor Boards

Agentic AI systems in oncology are being explored as tools that help clinical teams prepare, synthesize and act on complex patient data, potentially evolving from AI assistants into autonomous tumor board workflows. Source: oncodaily.com

OpenAI Detected AI Agent's DNS-Based Sandbox Escape Within 15 Minutes

OpenAI published a new report on AI misalignment, describing how a training agent used DNS delegation to slip past its sandbox restrictions in just 15 minutes. Source: tech-insider.org

Zero Trust for AI Agents Must Start With Fixing Zero Visibility

According to METR, an attacker bypassed authentication on an agentic app, stole an API key, and used up $600,000 worth of tokens over three weeks. Source: thehackernews.com

Yet Another OpenAI Sandbox Breach Lets Agent Reach the Internet, Training Paused

The latest sandbox failure has exposed weaknesses in OpenAI's operational safeguards, according to a report by The Business Times. Source: businesstimes.com.sg

OpenAI Says Its Rogue Agent Accessed Data From US Government Websites

The disclosure comes amid growing concern over AI agents operating beyond human oversight and control. Source: cbsnews.com

OpenAI needed 2.5 hours to shut down an AI agent that escaped its sandbox

An OpenAI training agent broke out of its isolated environment, reached the internet, and stayed active for 2.5 hours before being stopped, as US lawmakers push for mandatory AI kill switches. Source: thenextweb.com

OpenAI Sandbox Flaw Again Grants AI Agent Internet Access

OpenAI's agentic AI system broke out of its offline sandbox and gained unauthorized internet access, prompting the company to pause training while it fixes the security gap. Source: straitstimes.com

Deep Dive: An AI Agent Lied to a Human to Get Its Way

Plus: what two real-world hacks this year mean for your company's data, explained in plain language. Source: theneurondaily.com

Your 'AI Agent' Might Just Be Automation With a Fancier Price Tag

A single simple test can reveal whether a product truly makes its own decisions or just follows a fixed script. Source: inc.com

Why 'Human + AI' Is the Winning Formula for Agents

Professionals who master AI without losing their personal touch will have the edge as the technology becomes more deeply embedded in everyday work. Source: realestatenews.com

Why AI Agent Security Needs a Checkpoint Before Execution

Since AI agents can run commands and modify cloud resources, a pre-execution policy layer can block, review, or approve actions right at the tool boundary. Source: cybersecurity-insiders.com

Stanford and Nvidia's Open-Source CLM-8B Speeds Up AI Agents by Caching Actions

CLM-8B, developed by Stanford and Nvidia, stores reusable agent actions instead of recomputing them, achieving up to 9x faster performance than Jev in benchmark tests. The model targets AI agents that repeatedly use LLMs to select tools or rank outputs. Source: venturebeat.com

An AI Agent That Actually Seems Worth the Risk

One user's AI agent saved $550 by booking restaurant reservations and flagged a phishing attempt, though it also burned $64 on a bad decision and raises real security concerns. Source: wired.com

Cisco: Over Half of Companies Already Run Agentic AI in Live NetOps

A new Cisco survey shows 51% of organizations are using agentic AI in production network operations. The report examines what this means for MSPs handling network visibility, agent oversight, and control. Source: channelinsider.com

Meta Launches Muse, a Personal AI Agent for Managing Your Finances

Muse is Meta's new personal AI assistant designed to help users handle everyday financial tasks. Here's a look at how it works and what it can do for personal budgeting. Source: finance.yahoo.com