CTO Office / Head of Infra & Security Research @Bloomberg. Interested in mathematical techniques for software, systems, and network analysis at world scale.
New York City, NY, USA
Joined April 2010
- Tweets9K
- Following989
- Followers6.2K
- Likes1.3K
Pinned Tweet
Congratulations to the awardees of the Bloomberg Infrastructure and Security Ph.D. Fellowship 2025-2026. The 2026-2027 Call for Fellowship is now open! Apply by July 17th, 2026. bloomberg.com/company/storie…
Julien Vanegue retweeted
While GPT-5 & Claude Sonnet 4.5 can predict whether a program terminates nearly as well as top verification tool, ask them to *prove* it, and they break down. The researchers found a striking gap: models often know the answer, but can't show their work.
#EMNLP2026
(2/2)
Julien Vanegue retweeted
During Poster Session D at #EMNLP2026 today (11 AM CET), a group of researchers from Meta FAIR, UCL, ICL, HUJI, and Bloomberg, incl. @jvanegue, share research exposing a fundamental gap between #LLM semantic recognition and formal proof
bloom.bg/4kgZCgg
(1/2)
Glad to see Anthropic clearly communicate on the new scalable formal verification capabilities made possible by coding agents and frontier models. We can confirm this methodology is at work at @Bloomberg with Rocq, C++, Ocaml and other derived languages.
New on our Frontier Red Team blog: using Claude to formally verify software at large scale.
Formal verification can prove that code is free of certain bugs, but historically it has only been feasible for small codebases (or required a lot of expert time).
With AI models, however, this is beginning to change. In collaboration with @theoremlabs, we used Claude to prove security properties in over one hundred real-world open-source libraries over the span of a few weeks.
Read our research in full here: anthropic.com/research/forma…
We also solved the dual problem: reverse engineering the specification from the code, as shown by our paper “Fun2spec: Code Contract Synthesis At Scale” dl.acm.org/doi/pdf/10.1145/3… and Spec-Agent extending it for separation logic with fuzzing in the loop: arxiv.org/pdf/2605.27531
Funny thing, our harness was capable of solving that problem with open weight 32B models better than @AnthropicAI frontier models at a fraction of the cost. Frontier models are an expensive luxury, not a requirement.
I am en route to @dartmouth to give a special lecture on applications of program logic to security. Always a pleasure to see @sergeybratus and the whole team. Stopped in Boston on the way to visit @TuftsUniversity and get a cookie .
Consider submitting to the @poplconf workshop on theory and practice of static analysis. Deadline Oct 27th — popl27.sigplan.org/home/tpsa… — TPSA is a really great venue to meet top researchers in the field and tests your new ideas.
Using Opus 5.5 to develop Rocq proofs for program logic, I got hit multiple times by a warning Ive never seen before “Your session was flagged and your query blocked”. Definitely FP acknowledged by claude asking me to /bug report. The new guardrails need refinement @AnthropicAI
Julien Vanegue retweeted
I'm giving several talks next week in New York:
- Tuesday: NYC Math Ed & Communication seminar
- Friday: CUNY Logic Workshop
- Friday: CUNY Graduate Student Logic Seminar
In addition, I shall be recording four additional lectures in the ERGO studios for my Lectures on Infinity series.
Links below.
Afraid AI may put you out of a job? Worry not. Do like me and pick up a known undecidable task! Problem solved. 🫳🎤#nontermination
In my heavy usage of chatgpt 5.5 / 5.6 and claude code with opus 5 / 5.5 in the last few months, @OpenAI models have been superior to @Anthropic ones for mathematics. On a number of occasions, gpt was able to find issues in my drafts when opus had nothing to say.
Happy to see our paper "Mind the Heap: Preventing Capability Leakage Across CHERI Compartments" accepted at @acm_ccs 2026. I'll be in The Hague in November to support the amazing @s_hamedgh keeping the heap exploitation spirit alive on CHERI hardware. sigsac.org/ccs/CCS2026/progr…
This is joint work with @MauriceHanisch, now at @Caltech. Paper is open access and will appear online soon.
Julien Vanegue retweeted
I'm hiring an exceptional Offensive Security Researcher for my team at NVIDIA (Offensive Security Research - OSR).
Firmware, microcode, RISC-V, hypervisors, and shipping mitigations like HW CFI, Memory Tagging, and Pointer Masking from the ground up.
jobs.nvidia.com/careers?quer…
Julien Vanegue retweeted
If you would like to pursue a PhD at Imperial College London and work with me on blockchain and systems security-related topics, please contact me via my web site.
Julien Vanegue retweeted
I'm hiring a postdoc! Please share broadly with potential candidates.
I'm looking for someone with expertise in PL, FM, and/or networked systems who is excited about contributing to a research and teaching in a growing lab.
careers.epfl.ch/job/Lausanne…
Here is a really appealing process to PoC new research ideas:
1. Use ChatGPT voice assistant to brainstorm ideas and search related work throughout the day.
2. Your idea sticks? Generate a research plan and export in tex to GitHub.
3. Use Claude to iterate on implem & proof.