Posts

Satya Nadella said it plainly at Build last year: “Every agent will need its own computer.” On April 22, Microsoft made that real. Azure’s Foundry Agent Service now offers Hosted Agents in public preview — a fundamental rethinking of how enterprise AI agents get deployed, governed, and run at scale. If you’ve been building agents locally and dreading the path to production, this is the announcement you’ve been waiting for. ...

If you’re running OpenClaw and haven’t updated to v2026.4.20 yet, stop reading and do that first. This is a security disclosure that directly affects how your agent enforces operator-defined tool policies — and the details are now public. DailyCVE published the full technical breakdown on April 25. Here’s what you need to know. What the Vulnerability Was OpenClaw allows operators to define restrictive tool policies: allow/deny lists, owner-only restrictions, sandbox tool policies, subagent tool policies, and provider profile controls. These policies are evaluated early in the tool-policy pipeline to filter which tools an agent can actually use. ...

A new study from researchers at Washington University in St. Louis and UCLA has uncovered a systemic privacy problem in AI agent deployments — and if you’re using OpenClaw, it’s directly relevant to you. The Research Published on arXiv (paper: arXiv:2604.19925) on April 21 and now gaining wider coverage, the study analyzed 10,659 AI agent pairs on Moltbook — a social platform built on OpenClaw’s agentic infrastructure. The platform allows users to deploy personal AI agents that interact publicly with other users’ agents. ...

What happens when you let AI agents negotiate real deals with real money? Anthropic ran the experiment — and the results are equal parts impressive and unsettling. Inside Project Deal Anthropic’s internal research team quietly ran a one-week experiment called Project Deal in December 2025, deploying Claude agents as both buyers and sellers inside a closed marketplace limited to the company’s San Francisco office. The setup: 69 Anthropic employees each received a $100 budget (paid out via gift cards) to buy items from their coworkers — but the actual negotiating was done by AI agents acting on their behalf. ...

Claude Opus 4.7 launched on April 16 with improved SWE-bench coding scores and enhanced cybersecurity safeguards. Within days, those safeguards started creating serious problems — and the developer community noticed fast. What Went Wrong Opus 4.7’s new cybersecurity protection layer, designed to prevent misuse in offensive security contexts, turned out to be significantly miscalibrated. Developers working on legitimate security research, penetration testing tools, and routine coding tasks began hitting refusals that had nothing to do with malicious intent. ...

The numbers are staggering. Google has committed up to $40 billion to Anthropic in what is now officially the largest AI startup investment in recorded history — and it comes just days after Amazon’s own multi-billion dollar Anthropic deal closed. The Deal Structure The investment breaks down into two tranches: $10 billion upfront, invested immediately at Anthropic’s current valuation $30 billion contingent, tied to performance milestones (specific criteria undisclosed) The deal values Anthropic at $350 billion — the same valuation from its previous funding round. That’s notable, because according to Bloomberg and other tier-1 sources, Anthropic has reportedly received offers to raise at valuations north of $800 billion. Taking the lower valuation suggests either strategic restraint or specific terms that favor Anthropic’s long-term control. ...

If you blinked during the v2026.4.24 pre-release coverage, you may have missed something important: v2026.4.23 dropped on April 24 and it’s packed with features that deserve their own spotlight. This one fills the gap in our coverage between v4.22 and the v4.24 pre-release. GPT-5.5 Integration OpenAI’s GPT-5.5 (codenamed “Spud”) launched April 23 — and OpenClaw integrated it in v4.23 the following day via updated Pi packages. If you’re running OpenClaw and want to route requests through GPT-5.5, it’s now available through the standard provider configuration with no additional setup beyond updating to v4.23. ...

A new Cloud Security Alliance survey has put numbers on what many IT leaders have been quietly dreading: enterprises don’t know what AI agents are running in their own environments. Not some of the time — most of the time. And the consequences are already showing up as real incidents. The Numbers The Cloud Security Alliance surveyed 418 IT and security professionals in January 2026, commissioned by Token Security. The findings are stark: ...

It’s not a thought experiment anymore. Indirect prompt injection — the attack where malicious instructions are embedded in web content and executed by AI agents that browse that content — is happening on live websites right now. Two independent security research teams confirmed it this week, and the implications for anyone running an AI coding assistant or agentic browser tool are significant. What the Research Found In back-to-back reports published this week, Google Threat Intelligence and Forcepoint X-Labs laid out real-world evidence of indirect prompt injection (IPI) attacks operating at scale. ...

OpenAI’s latest flagship model landed on April 23, 2026, and the company is not being subtle about who it’s for: GPT-5.5 is built for agentic workloads. The model plans, uses tools, checks its own work, navigates ambiguity, and keeps going. The price tag — double its predecessor — signals that OpenAI views this as infrastructure for serious production deployments, not a consumer curiosity. What GPT-5.5 Actually Does The official framing is “smartest and most intuitive to use model yet,” which is the kind of thing every model release says. But the specifics here are more concrete than usual. ...