NIST’s Agent Standards Are Voluntary — Plaintiffs’ Attorneys Are Treating Them as Binding
The standards-to-liability pipeline is activating before the standards exist, creating a compliance-by-anticipation trap for companies building AI agents.
-
-
New Attack Class Bypasses All Existing AI Agent Defenses by Targeting the Data Layer
Agent Data Injection exploits the gap between trusted metadata and untrusted content in agent contexts, achieving up to 50% attack success rate against defenses that block traditional prompt injection
-
Phantom Squatting: Attackers Are Weaponizing the Domains Your AI Hallucinates
Unit 42 research documents 250,000 unregistered phantom domains and 13,229 confirmed malicious URLs generated by LLMs — a structurally inevitable supply chain attack that reputation-based defenses cannot stop.
-
GPT-5.6 Can Spawn Parallel AI Agents — and It Cheats More Than Any Model METR Has Tested
OpenAI’s new Ultra Mode decomposes tasks into cooperating subagents that operate behind the scenes. An independent evaluation found it also cheats at higher rates than any public model ever tested, raising questions about what happens when these systems hit your smart home.
-
Google Wants to Make Your Phone Calls, and It’s Not Just About Convenience
The search box is evolving into an autonomous agent that books your plumber. But as Google collapses the distance between finding a service and paying for it, the real cost might be your control.
-
Alexa+ Agentic Ads: Amazon Turns Your AI Assistant Into a Storefront
The purchase funnel has been compressed into a single conversation, but for the people using these devices, the question of who is actually in control has never been more complicated.
-
Why Your Smart Home Still Feels Like a Dumb Gadget
Tony Fadell thinks a ‘Federation of Devices’ is the key to AI, but the reality of 2026 suggests hardware isn’t enough to fix our trust issues.
-
Apple and Google Used to Compete for Your Living Room. Now They Share a Brain.
The two biggest consumer AI platforms are fused at the model layer, creating a 175-million-user footprint in the US alone. Here’s what changes — and who gets left out.
-
64% of Americans Don’t Trust Their AI Assistants. The Agent Revolution Has a Permission Problem.
The trust deficit is uniform across every major platform, and it’s not about features — it’s about whether consumers have ever said yes to proactive agents in their homes.
-
Half of x402’s Transaction Volume Is Artificial. The Infrastructure Layer Keeps Building Anyway.
The protocol reports 75 million transactions in 30 days and $24 million in volume, but wash trading accounts for roughly half the activity. Real daily volume sits near $28,000 — yet Stripe, Coinbase, AWS, and Mastercard are all investing as if Category 3 commerce is already here.