FuzzingLabs
@FuzzingLabs
Followers
8K
Following
728
Media
250
Statuses
747
Research-oriented Cybersecurity startup specializing in #fuzzing, Vulnerability Research & Offensive security on Mobile, Browser, AI/LLM, Network & Blockchain.
Paris
Joined August 2020
Generate videos in just a few seconds. Try Grok Imagine, free for a limited time.
1K
3K
11K
📊 Dotting (CFG Editing). Use --reduced or --only-entrypoint to create minimal graphs, and re-inject just the functions you need for analysis. Sol-azy lets you trim huge CFGs into something actionable.
1
0
0
⏪ Reverse Engine. Disassemble eBPF, track .rodata immediates, generate control flow graphs (CFGs), and even match instruction-level patterns.
1
0
0
🎣 Fetcher Module. Need the on-chain deployed bytecode of a program?. Sol-azy can fetch the actual .so from mainnet or custom RPCs.
1
0
0
⚙️ Build Engine (still in WIP). No need to fumble with anchor or cargo separately. Sol-azy can detect project type (and version of anchor/cargo) and compile .so artifacts for you, whether you're using Anchor or raw SBF.
1
0
0
📜 SAST Engine. Sol-azy uses a custom Starlark rule engine to scan the Rust AST and catch common issues statically. Write rules once, reuse them across audits.
1
0
1
Ever wanted a single tool to build, analyze (static analyzer), fetch, and reverse engineer Solana SBPF programs? 💻. We built Sol-azy, a modular CLI toolkit for security researchers: Let’s break it down 🧵
3
12
72
Finally!!!!! Look like we have some new targets to fuzz this summer 🔥.
0
2
23
Dive into the full methodology, raw numbers, and lessons learned in the blog ⬇️. 🔗
fuzzinglabs.com
We benchmarked 12 LLMs to find security flaws in code. Discover which AI models performed best and why overall accuracy remains a significant challenge.
0
0
1
Key takeaways:. - Training quality > parameter count.- Better orchestration + multi-agent strategies are the next frontier.- AI is a force-multiplier, not a silver bullet (yet)
1
0
0
⚠️ Reality check: every model produced 13-27 false positives. Human-in-the-loop review remains non-negotiable for the moment!
1
0
0
Community models hold their own! DeepHermes 3 lands 22.81 %, proof that open innovation matters.
1
0
1
Mid-tier open models punch above their weight:. - Mistral Medium 3 — 24 %.- LLaMA 4 Maverick — 23.53 %
1
0
0
Top 3 performers. 🥇 GPT-4.1 — 37.21 % score.🥈 Claude Sonnet 4 — 34.04 %.🥉 ChatGPT-4o — 31.91 %
1
0
0
How we tested ?. - Single-pass code review (no multi-agent tricks).- Languages: Python, Go, C++.- 6 vulns hunted: SQLi, XSS, Cmd Injection, Weak Crypto, Buffer Overflow, File Inclusion
1
0
0
Our @FuzzingLabs AI Team put leading LLM agents through a real-world vulnerability benchmark. Spoiler → even the best model still misses more than half the bugs! 🧵
4
2
15
RT @fede_intern: Great work by @FuzzingLabs that hopefully will be very useful for the @solana community. @0xMert_ and @aeyakovenko hope y….
0
2
0
🚀 We just released sol-azy on GitHub!. A modular CLI for static analysis & reverse engineering of #Solana sBPF programs — with disassembly, CFGs, and Starlark rule support. 🧵 👀 Docs: #Solana #RE #Security #Rust.
github.com
Sol-azy is a modular CLI toolchain for static analysis and reverse engineering of Solana sBPF programs - FuzzingLabs/sol-azy
2
16
57
Hardening checklist 🔐. - Upgrade to ≥ v0.3.15.- Bind to localhost or place behind VPN / reverse proxy.- Enforce auth (JWT, mTLS, API key).- Restrict egress & monitor logs for /api/create, /api/pull abuse.- Scan regularly (Shodan/Censys) for accidental exposure. 🔗 Full.
0
0
1
The CVEs you can’t ignore 🐞. 💣 CVE-2024-7773(RCE) — patch in v0.1.47.⚠️ CVE-2025-0317 (DoS) - patch in v0.3.15. Keeping legacy tags online = giving intruders shell access.
1
0
1