FeedCity

We simply don’t know to defend against these attacks. We have zero agentic AI systems that are secure against these attacks. Any AI that is working in an adversarial environment—and by this I mean that it may encounter untrusted training data or input—is vulnerable to prompt injection. It’s an existential problem that, near as I can tell, most people developing these technologies are just pretending isn’t there.

— Bruce Schneier

Tags: prompt-injection, security, generative-ai, bruce-schneier, ai, llms, ai-agents

It seems like every manufacturer of anything electrical that goes in the house wants to be part of the IoT story these days. Further, they all want their own app, which means you have to go to gazillions of bespoke software products to control your things. And they're

I lied in the title, well, sort of. Nobody explicitly asked for this but the signs were always there.

I finally updated my homepage! It’s not really flashy in any way, but it has a lot of links to what I’ve done over the last years. As part of it, I restored some of my “classic” Twitter and other social media threads, too. If you like my work, chances are you will find here something you enjoy.

Deconstructing the real costs of running AI inference at scale. My napkin math suggests the economics might be far more profitable than commonly claimed.

Piloting Claude for Chrome Two days ago I said: I strongly expect that the entire concept of an agentic browser extension is fatally flawed and cannot be built safely. Today Anthropic announced their own take on this pattern, implemented as an invite-only preview Chrome ex...

Today's links By all means, tread on those people: We know you love freedom, we just wish you'd share. Hey look at this: Delights to delectate. Object permanence: The right to bear cameras; GOP wants slavery for undocumented migrants; Telepresence Nazi-punching. Upcoming appe...

The cybersecurity community on Reddit responded in disbelief this month when a self-described Air National Guard member with top secret security clearance began questioning the arrangement they'd made with company called DSLRoot, which was paying $250 a month to plug a pair of laptops into the Redditor's high-speed Internet connection in the United States. This post examines the history and provenance of DSLRoot, one of the oldest "residential proxy" networks with origins in Russia and Eastern Europe.

Als er kamerleden voor de commissie Digitale Zaken gevonden moeten worden steekt bijna niemand in Den Haag zijn of haar hand op. Want iedereen weet, het kost je je carrière als je te digitaal bent. Je partij straft je af met een lullig positie op de lijst voor de volgende verkiezingen. Dit gebeurde harde werkers Hind Dekker en Lisa van Ginneken al eerder. En zo blijkt het ook dit jaar alweer te gaan.

Will Smith’s concert crowds are real, but AI is blurring the lines

Great piece from Andy Baio demonstrating quite how convoluted the usage ethics and backlash against generative AI has become.

Will Smith has been accused of using AI to misleadingly inflate the audience sizes of his recent tour. It looks like the audiences were real, but the combined usage of static-image-to-video models by his team with YouTube's ugly new compression experiments gave the resulting footage an uncanny valley effect that lead to widespread doubts over the veracity of the content.

Tags: andy-baio, ai, generative-ai, ai-ethics

Reverse Engineering ALL the Raspberry Pis

Today's links Enshittification, the audiobook (the Kickstarter): A disenshittified audiobook about enshittification. Hey look at this: Delights to delectate. Object permanence: Open source DRM?!; Sad Puppies lose big; Meritocratic delusion. Upcoming appearances: Where to find...

A deep dive into the convoluted and fascinating story of one of the most important keys on the keyboard. (First online appearance of an essay originally published in printed form in December 2023. 3,100 words. 35 photos.)

Agentic Browser Security: Indirect Prompt Injection in Perplexity Comet The security team from Brave took a look at Comet, the LLM-powered "agentic browser" extension from Perplexity, and unsurprisingly found security holes you can drive a truck through. The vulnerability w...

I'm fascinated by the unwillingness of organisations to name the "third party" to which they've attributed a breach. The initial reporting on the Allianz Life incident from last month makes no mention whatsoever of Salesforce, nor does any other statement I can find from

De hele korte versie: klassieke manieren om wetenschap en innovatie te stimuleren zijn erg ongeschikt voor disruptieve innovatie, ideeën die misschien helemaal niet gaan werken, maar als ze dat wel doen de wereld kunnen veranderen. En dat soort innovatie hebben we keihard nodig. De Duitse overheid heeft hiervoor een succesvol agentschap geschapen waar dit wel lukt, maar dat ging niet zonder slag of stoot. Nederland overweegt nu ook een “Agentschap voor Disruptieve Innovatie” op te richten.

Dismissing an idea because it doesn’t work in your head is doing a disservice to the idea. (Same for dismissing someone else’s idea because it doesn’t work in your head.) The only way to truly know if an idea works is to test it. The gap between an idea and reality is the w...

Saving Eorzea with as few keys as possible

W skrócie, wiele państw członkowskich UE, z Danią na czele, opowiada się za zmuszeniem WhatsApp/Signal/itp. do sprawdzania wszystkich naszych zdjęć i linków za pomocą sztucznej inteligencji. Jeśli sztuczna inteligencja ma jakiekolwiek wątpliwości, czy to może być pornografia dziecięca, Twoje zdjęcie, lokalizacja, numer telefonu i inne dane zostaną zgłoszone Europolowi i lokalnej policji. Co stanie się dalej, nie wiadomo, ale wakacje na Cyprze mogą skończyć się koniecznością tłumaczenia się ze swoich zdjęć na lokalnym komisariacie policji.

Static Sites with Python, uv, Caddy, and Docker Nik Kantar documents his Docker-based setup for building and deploying mostly static web sites in line-by-line detail. I found this really useful. The Dockerfile itself without comments is just 8 lines long: FROM ghcr.io/astral...