Tobias Pulls

Posted some technical details on DAITA v1 and v2 defenses together with @mullvadnet to celebrate the broad deployment of v2 🚀

ECH exposed a hard truth about privacy technology: you can win at the protocol layer and still lose at the deployment layer. The lesson from the ECH rollout is not just that a censor blocked it. It is that privacy can fail at the deployment layer even when the protocol is doing the right thing. I wrote about it here for @CenDemTech: cdt.org/insights/do-no… ECH's design goal is "do not stick out." If encrypted connections all look similar, they are harder to classify, monitor, and block. GREASE helped with that. It made ECH-shaped traffic common, so the syntax itself did not stand out. But that was not the whole story. Real deployments still produced a visible pattern. The issue was not the extension syntax. It was config update and recovery behavior. In practice, those recovery mechanics pushed clients toward a common visible outer name. That created a cheap classification handle. So ECH stopped sticking out at one layer and started sticking out at another. That is the interesting deployment lesson. Privacy is not just about cryptographic correctness. It is about operational indistinguishability too. Rollout paths, retry paths, and recovery paths matter. That is why signed ECH config updates are interesting. The point is not just "more crypto." The point is to remove the deployment constraint that created a stable fingerprint. This is the new draft with Dennis Jackson and Alessandro Ghedini: ietf.org/archive/id/dra… Interop work is here: github.com/grittygrease/e…

I’m making a bet with Filippo Valsorda that quantum computers won’t break ECC by 2029/2032, and (secondarily) that one version of ML-KEM will be de-standardized. I have loads of confidence in the former and little in the latter. I just like bets. dropbox.com/scl/fi/if3wfey…

Vinsterna är inte bara slöseri på skattemedel, de ger också elevens sämre inkomst under resten av livet. Hur kan M & Sd fortsätta försvara detta? efn.se/ny-forskning-e…

I wrote something: sockpuppet.org/blog/2026/03/3…

European countries should not rush into social media bans for children, human rights adviser Michael O’Flaherty told POLITICO. politico.eu/article/michae…

European countries should not rush into social media bans for children, human rights adviser Michael O’Flaherty told POLITICO. “A child has a right to receive information,” he said. 🔗 politico.eu/article/michae…

As co-comms chair of ICML 2026 (w @kgorman), I'm super proud of how transparent we've been able to be on all of the (bold!) decisions made. Thanks to all the organizers (esp PC chairs) for being aligned on this. The community deserves to understand these important decisions

I want to continue a bit on this subject, which (so far) I see very little concern about. There are vast stores of private data that we’ve built up in various places, including messaging apps. A real “killer app” for Gen AI is to ingest them and turn that data into revenue.

The wonderful thing about AI is that you no longer need the data. You need something to read the data and extract what you need. That “something” doesn’t need to expose data — it can be a local model or a TEE. The valuable product is the summarized data it extracts.

@_helenqu @cosmo_shirley @rdMorel @albertobietti @mikemccabe210 @PolymathicAI JEPA finds abstractions that enable prediction. This is how intelligence and science work: find abstract representations of reality that allow to make predictions.

If AI scientists are writing millions of papers, many of which are slop, and some of which are incremental progress, how would we identify the one or two which come up with an extremely productive new idea? In 1948, Shannon was one of hundreds of engineers at Bell Labs working on how to cleanly send voice signals over noisy copper wires. His paper sat in the same technical journal as reports on reducing static and building better filters. How would you recognize that he has come up with this very general framework for thinking about information and communication channels, which over the coming decades would have enormous use from domains as far apart as cryptography to genetics to quantum mechanics? It seems like it can take fields multiple decades to recognize the significance of unifying new concepts. Because it is on that time scale that the fruits of such general concepts lead to new discoveries across many different fields. We’ve managed to solve this peer review problem for human scientists (at least somewhat). Now we’ll need to do it at a much greater scale for the mass of AI science that will be thrown at us.

Who would have thought that academic research could be abused in such a terrible way 😉

Reading Dune. Frank Herbert was cooking.

"The purpose of our technology is to make all of you obsolete. Also, 10 to 25% chance it may kill the human race. Please deregulate in order to let us build this even faster, and don't let the government have any control over it." IS ONE HELL OF A PITCH

Others have noted this to, but the problem of age verification and “keeping ad supported networks profitable in the age of AI bots” are very similar problems.

What do you do when you get banned on British TV?

🇸🇪🇸🇪🇸🇪 €1.2 billion investment in Sweden’s digital infrastructure!

We’re about to add badly-designed age verification to the entire Internet: think how many extortion opportunities that will create.

The Signal developers are now hostile actors against the UK. As if they needed more of an ego boost.

@severinhacker the point of a PhD is not to get a PhD, it’s to do a PhD