We're doing an emergency CISO huddle next Tuesday to discuss ARTEX and autonomous attacks, and hear lessons from such incidents
Registration form for participation in the huddle:
forms.gle/AEL81hrNUbZ6qNcf9
- Tuesday, November 17, 2026
- 8 am Pacific / 11 am Eastern / 5 pm CET.
🎤 We’re excited to host a panel on “Democratizing Software Understanding: Are We There Yet?”
It’s an honor to have @sergeybratus , @insu_yun, Johannes Kinder and Sébastien Bardin join us for this discussion moderated by @mahal0z.
If you’re around, come join us!
Rethinking computing in the AI-coding era has been liberating. My accelerated intro class now has an assignment done entirely using Z3. Take three prior assignments and redo them using constraints and optimization; compare and contrast. See:
cs.brown.edu/courses/csci019…
Both @OpenAI and @AnthropicAI have started flagging #Rocq proofs generation as problematic. Someone needs to explain them that generating .v files has never hacked anyone.
30 years ago today, Phrack 49 published Aleph One’s “Smashing the Stack for Fun and Profit”.
A legendary paper that inspired generations of hackers and helped shape modern exploit development.
Still an essential reading.
phrack.org/issues/49/smashin…
Phrack 73 is coming soon!
[1/3] turns out you can teach a 35b model to weight-space program better than opus o_0
earlier this week, @rikiyatakehi and @jayden_teoh_ showed that astra can manually write neural networks to solve atari games, one weight at a time. they call this weight-space programming.
on the @trajectorylabs platform, @o__jerry__o and I post-trained 27b and 35b-a3b qwen models to write networks for atari games. we curated WeightProgramBench, a set of 322 simple games to teach models to manually write neural networks.
we combined 51 atari-style games with 266 games adapted from crisp-game-lib for training, holding out the 5 original minatar games (freeway, breakout, asterix, seaquest, space invaders).
on freeway, qwen-27b climbed from averaging 21.0 to 47.1 points **without ever training on freeway**, but still not quite at astra’s 63.9 points.
The School of Computing at UT Austin will be hiring aggressively, aiming for 50 new tenure-track faculty over the next 5 years, and many instructional (professional-track) faculty.
Job ads are posted for all 3 departments:
I was planning on holding off on this announcement until the project was farther along, but I'm excited to announce that I've been working on a new programming language called Phil, and the team at @Higherdotfun_ is supporting its development.
Today we begin the second and final day of vibecheck. The most exciting part of this is judging! Yesterday I explained to a contestant (not Holmes, someone else) that his project would be evaluated on 1) coolness and 2) meaningfulness of proofs. He asked how we would evaluate 2. I said we’d read his proofs. He was like, “fuck, you’ll be the first person to do that!” LOL. So, without a molecule of criticism, what does it mean to run Slack in Lean? Is the Lean code interpreting JavaScript? Emitting JavaScript? Emitting WASM? Something else? We’re gonna find out! Super super interesting space, and we will report back on the wide diversity of outcomes we encounter, both good and bad!
"verified Slack alternative" is me!
It feels totally sci-fi to me that you can build usable Android and iOS apps, in Lean, that prove most of their important properties.
The big assumptions are: "the OS scrolls and tells us about clicks and taps correctly", "libsodium is correct" and "SQLite is correct".
Most important properties can be written in Lean and proven.
Periodic re-up of @rossjanderson (RIP)'s definition of "security engineering"
"Security engineering is about building systems to remain dependable in the face of malice, error, or mischance. As a discipline, it focuses on the tools, processes, and methods needed to design, implement, and test complete systems, and to adapt existing systems as their environment evolves."
cl.cam.ac.uk/archive/rja14/P…
I’m very, very excited to announce that @dartmouth professor and former @DARPA pm @sergeybratus will be giving the opening remarks tomorrow morning to kick off Vibecheck, the formal methods hackathon!
About a decade ago, I went to Sergey’s office and told him I wanted to be a hacker and I loved math. He told me to drop my EE major and switch to pure mathematics — the best foundation for a real hacker. He also gave me an enormous book by Dennis Yurichev on reverse engineering, which meant that hilariously, I was reading about linker/loader internals before I’d ever even tried building a compiler. In fact I learned about C reverse engineering before I learned C. (Not claiming I was any good at it, just that my sequence of exposures to the topics was comedic ..)
Sergey is known for many things but maybe the most fundamental is the co-invention of LangSec, the field of cybersecurity which posits that many — well, most — exploits are in fact parser vulnerabilities, and therefore security should be approached from the perspective of formal language theory. This insight has proved incredibly prescient, predicting vulnerabilities like Pegasus, or more recently the incredible node exploit. But it’s also prescient in another way. More philosophically, we have these language models generating code, and no good way to parse their outputs to determine if the code is safe or not. If you’ll allow me the slight stretch of the formalism I think this is basically a LangSec issue .. and the solution, AI enabled formal methods, is basically a parser innovation. Modulo how do you find specs, which of course is a more nuanced conversation involving HCI and more.
We will have many experts on the floor at vibecheck from folks like @mathematics_inc@theoremlabs@workersio@lanyon_ai and more. But I couldn’t think of anyone I’d rather have open the event than Sergey. See you all at 10am @thegp !
We are honored and excited to have Dr. @sergeybratus give the opening keynote at SURE 2026: "Defining the discipline of Software Understanding in the age of AI"
Join us at SURE on November 19, 2026: sure-workshop.org/program
Full program coming soon!
Webp, now Brotli, recently introduced complex formats that eschewed formal mechanized definition and code generation in favor of hand-crafted imperative implementations. Billed as "modern", backed by supposedly superior development processes. Turned out as usual.
"Your browser extensions or network settings have blocked the security verification process required by [a news website] To resolve this, [turn off all defensive filters]". -- No thank you. I hope your news are worth what you're paying to protect them from readers like me.
After months of getting things done w/ Claude Code, I've spent ~5 days programming by hand in Lean and a liquid typed language and it's…SO GREAT? I love living in a world where I can get 1000s of mindless LOC in a minute, but also spend an hour on 10 gorgeous handcrafted lines.