Program analysis. Reverse engineering. Backdoor detection. Head of Research @RevEng_AI. Creator of @VulHuntRE.
United Kingdom
Joined December 2017
- Tweets737
- Following651
- Followers720
- Likes3K
Sam Thomas retweeted
Huge findings from @yeggorv💪
1/3 We published six advisories for vulnerabilities in NVIDIA GPU kernel modules and NVIDIA DGX Spark device firmware that we discovered at the beginning of the year. Taken together, these vulnerabilities could allow an attacker to escalate privileges from an unprivileged user to the kernel, and from the kernel to Standalone MM on a DGX Spark device, using only the NVIDIA stack!
Sam Thomas retweeted
Excited to see our paper, “A Hybrid Static-Dynamic Analysis for Identifying BYOVD Vulnerabilities,” now public! It introduces dynamic sandbox detection and IOCTLance features, leading to 13 reported vulnerabilities and 5 CVEs.
sciencedirect.com/science/ar…
Sam Thomas retweeted
📢 IDA 9.5 is here: more platforms, more agentic power.
→ 3 new decompilers: Android Dalvik, Infineon TriCore and Qualcomm Hexagon
→ IDA MCP and Assist for AI-driven reverse engineering
→ Recursive decompilation across call graphs
→ IDA Home now includes 11 cloud decompilers instead of 2, at the same price
→ And much, much more
It’s one of our biggest releases yet. Active license holders can download it now from the Download Center.
hex-rays.com/blog/ida-9.5-re…
Sam Thomas retweeted
Excited to announce that I'll be keynoting at Black Hat Europe this year in London! I'll be talking about the distinction between AI Security and AI Safety, and the co-option and conflation of security and safety methodologies in AI that are leading us down a precarious path.
This quoted post is unavailable.
Sam Thomas retweeted
understanding and benchmarking vulnerability detection capabilities is an interesting challenge and I can recommend @niklas2484's recently published PhD thesis on that topic.
it includes "Top Score on the Wrong Exam", one of my favorite (pre-LLM) ml-for-vuln-detection papers.
Sam Thomas retweeted
A new blog post "Ghidra + CBMC = Assembly Model Checker" philipzucker.com/asmc_cpp/ A fairly small tool can lift assembly to CBMC's GOTO IR enabling simple cross C-assembly model checking for many ISAs (Ghidra supports 20+ github.com/NationalSecurityA… ) #ghidra #assembly #verification
Sam Thomas retweeted
Wrote some slides on symbolic exploit search / synthesis... kmicinski.com/ai-security-co…
Sam Thomas retweeted
DecBench is growing fast, but I want to remind people that I manually verify results and submissions on the site. That means two things: 1) if you see results elsewhere and not on DecBench.com, I can't speak to their quality. 2) Result updates can be slower, so please don't spam 👍
Sam Thomas retweeted
I am recruiting a postdoc to work with me here in Paris on fuzzing and backdoor detection!
➡️ Do not hesitate to apply or share the offer!
🚨 Postdoc opening (24 months) in software security & fuzzing!
Join our @BinsecTool team @CEA_List to work on smarter fuzzing for supply‑chain security
📍 Paris‑Saclay 🇫🇷
🔗 Apply: binsec.github.io/jobs/open/2…
#Postdoc #Cybersecurity #Fuzzing #BinaryAnalysis
I'm pleased to announce a new release of the #idalib #Rust bindings for @HexRaysSA IDA Pro! This release provides new APIs and bug fixes. Thanks to new and existing contributors @mcbbugu, Amaan Qureshi, and @0xdea for their work on this release.
github.com/idalib-rs/idalib.…
Sam Thomas retweeted
A new decompiler approach enters the ring!
today, we @RevEng_AI released the next generation of our decompiler: Ventris
the new model comes with improved struct layout recovery, better PE support, and more.
on decbench, we beat all other available decompilers in structure recovery, and achieve #2 rank overall
Sam Thomas retweeted
Super interesting work, and improving on structure recovery is really impressive; that's a seriously hard problem that has had a lot of previous work
today, we @RevEng_AI released the next generation of our decompiler: Ventris
the new model comes with improved struct layout recovery, better PE support, and more.
on decbench, we beat all other available decompilers in structure recovery, and achieve #2 rank overall
Sam Thomas retweeted
it's great that with decbench, we have a high-quality public benchmark.
yet, decompiler evaluation remains hard, and I see two primary challenges right now:
1. high-quality, real-world-like, fully private data. Frontier LLMs have seen virtually every open source function in existence and leakage is not fully avoidable, something we also cannot rule out for our model.
2. finding a better proxy for semantic fidelity. A program can recompile, but do something entirely different. A program can also recompile, have a totally different byte-level structure from the groundtruth, and still do the same job. Dynamic-execution based attestation is tough to get correct at scale, but might be a better signal in the long run.
Sam Thomas retweeted
we submitted and verified our results via the public site: decbench.com/leaderboard/?da…
(`normalize` as there is no arch-specific view right now)
it's cool to see that our {static context collection -> one-shot decompilation} approach can compete with fully agentic frontier LLMs
Sam Thomas retweeted
today, we @RevEng_AI released the next generation of our decompiler: Ventris
the new model comes with improved struct layout recovery, better PE support, and more.
on decbench, we beat all other available decompilers in structure recovery, and achieve #2 rank overall
Sam Thomas retweeted
I've shared the list of malicious models detected by MLTracer, along with a systematic analysis of techniques for evading existing model scanners, in the following GitHub repo.
github.com/binarly-io/ai-sup…
In short, the AI supply chain security field still likely has a lot of room for improvement.
🚨 Our research found that multiple major Hugging Face scanners missed more than 20% of malicious model files.
We systematized the 21 static-scanner evasion techniques behind these misses. Most of these were based on known techniques or concepts, highlighting an inherent limitation of static pattern matching: it is practically infeasible to cover every concrete malicious variant.
binarly.io/blog/malicious-mo…
Sam Thomas retweeted
"Software protection is futile, AI will break it", they said.
But, does it? how?
To shine some light read about Rémy Salim's experiments in "Defeating AI-Assisted Reverse Engineering"
Where is the fair play, Claude?
blog.quarkslab.com/defeating…
Sam Thomas retweeted
Schrödinger’s TOCTOU — a bug class where a struct’s size mod 16 decides if you’re exploitable, where -O2 saves you and -O3 kills you, where a compiler upgrade introduces the bug, and a downgrade… introduces it too. Invisible, intractable, everywhere.
github.com/xoreaxeaxeax/schr…
Sam Thomas retweeted
Static Devirtualization of Tencent VM. There are many others with similar results to ours, we have had the VM devirtualized for several months now. Others are posting full devirt bins, so we feel it is ok to release our research now.
back.engineering/blog/31/07/…
I'm pleased to announce a new release of the #idalib #Rust bindings for @HexRaysSA IDA Pro! This release adds compatibility with latest SDK (v9.4), and a couple of bug fixes. Thanks to @0xdea and @yeggorv for their contributions.
github.com/idalib-rs/idalib.…