New paper: In a small study with 32 people who were learning science from an AI tutor that was confidently wrong, the participants took over around two of three of its false informations, and they felt almost as sure about them as about the true facts.
Their trust in the AI did barely drop, what suggests that sometimes it's not the AI who bends to the user, but the user who bends to the AI.
The technical landscape gets more complicated every day, so digital products need to state their cornerstones clearly.
Here are ours at PrunePad: a card for each of our six AI agents with what it reads, remembers and won't do, and a list of what we don't claim.
If you're a DPO, look after IT for clients or are just curious: what's missing from a page like this before you'd let an agent into the chat of your team?
Excited to see W3C taking on how AI agents are changing the web.
Its new discussion paper argues that agents should follow the same rules on security, consent, accessibility and openness that protect the rest of us, and that's exactly the conversation we need to be having right now.
It will be presented at TPAC on 29 Oct 2026.
Which action at work would you never let an agent take without asking you?
https://www.w3.org/2026/10/w3c-and-state-of-web-in-ai-era.html