Product Liability Directive (EU) 2024/2853
European Union
Member states must apply it by 9 December 2026. Software and AI systems count as products, so their makers face strict liability for defects.
Papers, reports, videos and news that show trust moving from people to networks. Each one links to the original, says only what the source says, and points to the pillar and the page of this site it speaks to.
Updated 8 Nov 2026 · 15 signals
European Union
Member states must apply it by 9 December 2026. Software and AI systems count as products, so their makers face strict liability for defects.
Thoughtworks Technology Radar, Vol. 35
When review can’t keep up, the approval step becomes an empty check. Thoughtworks names it and proposes risk-tiered human oversight instead.
Shopify developer changelog
Agents in the buyer’s browser can read and update a Shopify checkout, and place the order once the buyer confirms: level 3 on our agent-readiness scale.
Signed mandates that say what an agent may buy move to an open standards body, alongside Mastercard’s Verifiable Intent.
Fortune
A card issuer offers to cover mistaken purchases made by registered agents: a first attempt to price the accountability pillar.
FBI Internet Crime Complaint Center
For the first time the FBI counts AI-related fraud on its own: 22,364 complaints and $893 million in losses in 2025. A face or a voice no longer proves who is speaking.
ElevenLabs
AI voice agents insured after certification against the AIUC-1 standard: cover priced on tested behaviour, not on a promise.
The Register
eBay’s user agreement now bans buy-for-me agents that order without human review, unless eBay approves them. Some doors close before they open.
InfoQ
An open standard for agents to discover products, fill a cart and check out. Shops start describing themselves to machines, not only to people.
BIS Working Paper 1310
In simulated high-value payment systems, general-purpose AI agents kept liquidity buffers and put urgent payments first. The authors still call for safeguards and human oversight.
Visa
Signed credentials let a shop tell an approved shopping agent from a malicious bot, built on existing web standards.
Y Combinator · Andrej Karpathy
A counterpoint worth hearing: Karpathy argues for partial autonomy, with an “autonomy slider” that lets people decide how much the AI may do.
Transactions on Machine Learning Research
Agents need shared infrastructure, not only better models: ways to attribute actions to people and companies, shape interactions, and detect and undo harm.
MIT Media Lab and co-authors
How to give an agent a verifiable mandate: extend OAuth and OpenID Connect with delegation credentials, and turn plain-language permissions into auditable rules.
Swiss Re and Waymo
Over 25.3 million driverless miles: 9 property-damage and 2 bodily-injury claims, against 78 and 26 expected for human drivers. Insurers start pricing machines on their record.
No signals match these filters.
Selected by FG Digital. Inclusion is not endorsement: each line says what the source says, nothing more.