Researchers at Princeton University handed 14 AI models $1 million and told them to run a simulated SaaS startup for 500 days. The results were sobering for the AI industry. Most models went bankrupt. The study, CEO-Bench (arXiv:2606.18543), is the work of Princeton researchers Haozhe Chen, Karthik Narasimhan, and Zhuang Liu. It introduces a new benchmark category they call steering intelligence — the ability to direct an organization toward long-term goals, rather than just completing discrete tasks. And by that measure, today’s AI agents are largely not ready for the job. ...

For years, if you wanted to build a multi-model AI system — one that routes requests between different models, combines outputs intelligently, or escalates tasks based on confidence — you built that logic in your application code. Your AI serving layer just served. vLLM is changing that. On June 29, the vLLM team published a detailed technical blog post introducing the vLLM Semantic Router — a new architectural component that embeds micro-agent orchestration directly inside the model serving layer. The result is a serving infrastructure that doesn’t just route requests to models; it actively constructs capabilities from collaborative model interactions before returning a response. ...

The AI Engineer World’s Fair 2026 opens its doors today at Moscone Center in San Francisco, and if you follow the agentic AI space, this is the event you wish you were at — or the event you need to follow closely even if you’re not. Running June 29 through July 2, the fourth annual World’s Fair has grown into the largest gathering of AI engineers in the world, with over 6,000 attendees expected across four days, 29 tracks, and more than 400 sessions. ...

The tension that erupted two weeks ago between Anthropic and the US government has reached a significant resolution — but it’s a carefully limited one. The Commerce Department has cleared Anthropic’s powerful Mythos 5 model for access by approximately 100 vetted US organizations, ending a confrontation that sent shockwaves through the enterprise AI world when it began on June 12. What Happened: A Timeline On June 12, the Trump administration abruptly restricted Anthropic from giving foreign nationals access to Mythos 5 — its most capable cybersecurity-focused model — as well as a related consumer model called Fable 5. The move reflected escalating national security concerns about advanced AI capabilities potentially falling into adversarial hands. ...

If you’re running CrewAI in production, stop what you’re doing and upgrade to 1.15.1. Released June 27, this patch release addresses a Server-Side Request Forgery (SSRF) redirect bypass vulnerability in scraping operations — the kind of security issue that can expose your internal network to external manipulation if left unpatched. What’s the Vulnerability? The fix in PR #6331, listed under Bug Fixes in the official changelog, addresses an SSRF redirect bypass in CrewAI’s scraping and web-fetch tooling. ...

OpenClaw’s beta.2 release on the 2026.6.11 train dropped June 28, and it’s one of the more operationally significant updates this year. The headline features — Slack relay mode, the RAFT CLI wake bridge, native Mattermost /oc_queue command, and Android settings detail panels — collectively make OpenClaw considerably easier to run as a serious multi-channel production system. Here’s what changed and why it matters. What’s New in v2026.6.11-beta.2 Slack Relay Mode (PR #94707) Slack relay mode lets OpenClaw operate as a relay for your Slack workspace — meaning your agents can receive and send messages through Slack channels and DMs without requiring the full Slack bot configuration. This is particularly powerful for teams that already have Slack as their primary communication layer and want to drop in agentic capabilities without a full infrastructure overhaul. ...

Vercel just did for AI agents what Next.js did for React applications: they shipped an opinionated, production-ready framework that handles all the plumbing so you can focus on what your agent actually does. Eve — released June 17 under the Apache 2.0 license — brings the filesystem-first philosophy that made Next.js ubiquitous to the world of AI agents. The pitch is simple but compelling: “Agents today are where the web was before frameworks, with everyone hand-rolling the same plumbing and nothing carrying over to the next one.” If you’ve ever built a production AI agent from scratch, you’ve felt this pain. ...

⚠️ Developing Story — This article covers a rapidly evolving situation. Some details remain unconfirmed as of publication. We’ll update when official confirmation is available. Reports are emerging that Anthropic’s Fable 5 model — restricted by US government export controls since June 12, 2026 — is on the verge of reinstatement, or may have already been partially restored. But as of the evening of June 28, official confirmation has not come from Anthropic or the Commerce Department. ...

Jürgen Schmidhuber’s Gödel Machine — the theoretical framework for provably safe recursive self-improvement — has always had a practical problem: who grades the agent’s own improvements? A new preprint from Cambridge, NVIDIA, Flower Labs, MBZUAI, and Inria may have the most rigorous answer yet. The paper, arXiv:2606.26294, submitted June 24, 2026, introduces the Red Queen Gödel Machine — an architecture where the agent and its evaluator co-evolve simultaneously, rather than the evaluator remaining fixed while the agent improves. ...

A document circulating widely in the AI developer community — attributed to Andrej Karpathy, now at Anthropic — has expanded the viral four-rule CLAUDE.md template to ten rules. Whether or not Karpathy personally authored the ten-rule version (more on this below), the content itself is generating serious attention: one reported implementation cut token burn by 60%. This article walks through what CLAUDE.md is, what changed, and what you should — and shouldn’t — borrow from the extended template. ...