★ Featured Essay
Who and How
A new UN initiative on trust in AI agents opens by naming two questions that earlier trust efforts kept collapsing into one. That, more than anything it has built yet, is worth being glad about.
Read Essay →Thought Leadership
Original thinking at the intersection of research, practice, and the messy reality of how people actually work with intelligent systems.
★ Featured Essay
A new UN initiative on trust in AI agents opens by naming two questions that earlier trust efforts kept collapsing into one. That, more than anything it has built yet, is worth being glad about.
Read Essay →All Writing
Six Thousand Five Hundred Packages
IBM and Red Hat announced new "trust infrastructure" for open source software. Signed binaries, certified dependencies, an AI remediation engine, a clearinghouse for coordinating patches. Nine paragraphs, and not one mention of the person deciding whether to trust any of it.
→Thirty-One Seconds
Security researchers documented an AI agent that ran an entire extortion campaign without a human in the loop. What they could not document is whether the machine understood any of it.
→Three-Quarters Agreed
A week of AI trust-gap reporting converged on the same fix: more testing, more monitoring, more governance. The numbers inside one of those reports suggest the fix was never the thing missing.
→The Goats Were Never Agentic
A Microsoft researcher built a working computer out of Age of Empires II goats to prove a point about language. The point applies just as well to the gap between what people understand about AI and what they trust it to do.
→The Default
When the output looks finished, the question of how it got there tends not to get asked. What AI design homogeneity reveals about trust, agency, and the questions defaults foreclose.
→Recon This
A low-skill attacker. Fourteen breached companies. One prompt. What the session logs reveal about trust in agentic AI — and what guardrails were never built to catch.
→What the Camera Cannot See
Designers are building garments that defeat facial recognition. They are working from the same map computer vision researchers have been studying for years.
→The Signal in the Noise
An AI caught something the doctors missed. What that moment reveals about trust, comprehension, and what happens when a flag arrives without an explanation.
→The Day the Caps Kicked In
What a lawsuit against Anthropic reveals about trust, pricing, and what users actually understood when they subscribed.
→The Process Is the Point
When a government pulls an AI system offline without showing its work, it hasn't made anyone safer. It's just made the problem invisible.
→Crisis Adoption in the Wild
A $16.9 million contract, a financial emergency, and what happens when an entire university system adopts AI before it knows what it needs.
→Still Waiting on the Bill
When a tool amplifies capability in both directions, releasing it is not a technical decision. It is a trust decision.
→The Day the Bill Arrives
What happens after the crisis adoption — when the tab comes due and no one has established what any of it was worth.
→AI Should Happen With Us
On accountability, agency, and what it actually looks like to work alongside AI — not beneath it.
→The Crisis Adoption Problem
Why we reach for AI hardest when we can least afford to get it wrong — and what that costs us.
→The Bandwagon Problem
Why "everyone is doing it" is the most dangerous reason to adopt AI — and the most common one.
→The Infinite Team Problem
Meta just launched an AI agent designed to replace your customer service team. My research suggests we should be asking who's still in the loop — and what happens when no one is.
→Trust Without Understanding
82% of AI deployments fail to move beyond proof of concept. The reason isn't technology. It's cognition.
→Can AI Replicate Authenticity?
A math rock band from Quebec, a viral performance, and what they reveal about human intent in the age of AI.
→