A ‘Verified’ Plugin Can Run Any Code It Wants on Four AI Coding Agents. Two Vendors Won’t Fix It.
Plugin4Shell defeats the SHA pinning that was supposed to lock AI coding agents to reviewed plugin code. Claude Code and Codex patched it; Copilot and Gemini CLI did not.
OpenAI and Google Just Shipped Voice Agents Five Days Apart. They Disagree on Where the Brain Goes.
GPT-Live-1 and Gemini 3.8 Live landed five days apart. Voice quality has converged, so the real 2026 voice-agent choice is architecture, latency, and cost at concurrency.
The House passed the Ratepayer Protection Act 417-3 to stop AI data centers from pushing grid costs onto households. The catch is in the mechanism, and in what state regulators were already doing.
Claude’s Agents Can Now Message Each Other. ‘Sharing Context’ Is the One Thing They Won’t Do.
The headlines say Claude's parallel sessions now share context across terminals. Anthropic's own docs say they pass text notes and nothing else. That gap is the whole design decision, and the...
The AI Label You Can See Won’t Pass December 2. The One That Will Is Invisible.
A visible AI label meets half of Article 50. The machine-readable half, where C2PA, SynthID, and IPTC fail in opposite ways, is the part that decides whether you pass the December 2 deadline.
You Have Until December 2 to Watermark Your AI Content for the EU. A Visible Label Isn’t Enough.
The EU AI Act requires machine-readable marking on AI-generated content by December 2, 2026. A visible caption satisfies a different rule. Here's what actually qualifies, why C2PA metadata alone gets...
