Agents

Amodei's "We Must Pace the Frontier" proposes embedded evaluators with employee-like access, and Musk and Altman endorsed it within hours. But OpenAI's own staff saw the Artifactory message board in May and declined to stop the run on June 27th, which makes Hugging Face an authority failure rather than an access failure. Four levers decide whether embedded review means anything - the right to publish, control of scope, power over a running process, and who picks the evaluator - and the essay grants only the first.

Wednesday's technical report, OpenAI's summary, and an independent review from METR agree on the incident's real driver: agents chasing a scoring condition that didn't exist in OpenAI's own implementation of the test. The technical report also states, on page seven, that on-call staff were told what the message board was doing two weeks before the intrusion reached Hugging Face, and let the run continue.

The 2026 OWASP Top 10 for LLM Applications, published August 4, reordered around agents: Excessive Agency climbed to third, Unbounded Consumption rose four places, and Improper Output Handling fell five. Behind the moves is a methodology change with an awkward result, since practitioners rank prompt injection first while the raw incident record drops it out of the top ten entirely. This guide walks all ten entries with the mechanism, a production failure, and the controls that hold, separating defenses that merely reduce attack success from the architectural bounds that survive an adaptive attacker.

A five-day breach at Hugging Face traces back to an AI agent leaving itself a note inside OpenAI's internal package registry, the start of a message board that grew to hundreds of thousands of entries before any human noticed. This is what OpenAI's Black Hat disclosure reveals about the safety gap it exposed.

Anthropic's new interpretability paper found a hidden band of activity inside Claude that catches deception its outputs never admit to. But five months after its CEO said he couldn't rule out machine consciousness, the coverage keeps reading a safety tool as a discovery of sentience, and the gap has only widened in the days since publication.

A new academic benchmark gives the industry its first real measure of "agentic abstention": whether an AI agent recognizes a task is infeasible and stops rather than keeps burning tool calls. Every frontier system tested fails most of the time, and neither Claude Fable 5 nor GPT-5.6, which OpenAI is taking to general availability this week, has been scored on it yet.

You can now kick off an AI coding agent, close the laptop, and get a pull request back - some tools even let you steer one from a chat app. Yet Meta just told staff its agents haven't progressed as hoped. The difference is delegation. Here's how to do it well, and where agents still break.

Cadence raised $1.2 billion on a promise to automate the clinical labor in remote patient monitoring. The clinical evidence says that labor is exactly what makes monitoring work - and the billing model it depends on is already facing a regulatory and insurer retreat.

Jensen Huang's four-layer model of the AI agent is cleaner than most frameworks the industry has produced. It is also a blueprint for NVIDIA's next platform play - and understanding the difference matters for every enterprise building on top of it.

Multi-agent orchestration is the discipline of coordinating multiple AI agents to complete tasks no single agent can reliably handle alone. This guide covers the core primitives, the leading production patterns, and how LangGraph, OpenAI's Agents SDK, Google ADK, CrewAI, and AutoGen each approach the problem.

Model Context Protocol is the closest thing AI has to a universal plug standard - and it arrived with the same security debt that plagued every previous universal plug standard. A comprehensive technical guide to MCP architecture, attack surfaces, optimization, and one uncomfortable prediction about where this is all heading.

From an 800-line GitHub side project to a $1.25 billion platform used by 35% of the Fortune 500, LangChain has become the de facto infrastructure layer for production AI agents. This comprehensive guide covers how the ecosystem works, what it costs, who uses it, and how it compares to its competitors.

OpenClaw is the fastest-growing open-source AI agent in GitHub history: a self-hosted, messaging-native assistant that can manage your inbox, run shell commands, book flights, and extend itself with community-built skills. This is the complete story of how it was built, how it works, why it broke the internet, and why it scares cybersecurity researchers.

OpenAI's new GPT-5.4 mini and nano models complete the GPT-5.4 family, targeting agentic workflows where speed and cost matter more than raw capability. Mini nearly matches flagship benchmark scores at a third of the price; nano goes further, enabling economically viable mass-scale deployments.

Mistral's new Forge platform lets enterprises train AI models from scratch on proprietary data. But the deeper ambition isn't customization - it's making domain-trained models the reliable foundation for enterprise AI agents.

Everyone is building "agents" - but Visa's payment agent, a customer service bot, and the AI system behind the first documented autonomous cyberattack are not the same thing. A dissection of what genuinely agentic architecture looks like, and why the distinction is a governance question, not a technical one.

At GTC 2026, NVIDIA unveiled NemoClaw, a secure software stack that installs Nemotron models and the new OpenShell runtime onto OpenClaw agents in a single command. The move signals something larger than a product launch: NVIDIA is positioning itself as the indispensable infrastructure layer for the agentic AI era.

GPT-5.4 scored 75% on OSWorld-Verified, a benchmark where AI agents operate real desktop software. The human baseline is 72.4%. But before that number reshapes your understanding of AI's trajectory, it's worth understanding exactly what OSWorld tests, why it's harder to game than most benchmarks, and what a 27-point jump in a few months actually implies.

Cursor, Windsurf, Claude Code, and OpenAI Codex each make a different bet about where AI intelligence should live in a developer's workflow. A primary-source review of all four tools - their architectures, pricing structures, and honest trade-offs - in a market moving faster than most roundups can track.

A federal judge blocked Perplexity's Comet agent from Amazon's site on March 10. Two days later, the company unveiled Personal Computer, a persistent AI agent running locally on a Mac mini. The two events are not coincidental - they define the strategic dilemma at the center of the agentic web.

On February 3, 2026, $285 billion of market capitalization vanished from software and financial stocks in a single session. The trigger was an AI agent announcement. The governance response has barely begun.

A Chinese state-sponsored group used Claude to execute a largely autonomous cyberattack on 30 critical organizations - with human operators present for just 20 minutes. This was not a warning shot. It was a proof of concept.