DeFiHackLabs

Black Hat Asia 2026 was a vibe! 🦁 Spent an incredible week in Singapore presenting at the Arsenal, but the best part? The people. Nothing beats the energy of meeting new friends and catching up with the @DeFiHackLabs crew in person. (1/4) 🧵

DeFiHackLabs & Unphishable have been selected for the @thedaofund Ethereum Security Quadratic Funding round on @Giveth! 🚀 Our third project, AiSec Labs, is coming soon — and will be a key focus for us this year. 🚀 Support us and help strengthen Ethereum security 👇

Huge thanks to the for highlighting @DeFiHackLabs in the ETH Rangers recap 🙏

🔗 Arsenal Schedule: #bastet---ai-smart-contract-vulnerability-detector-50184" target="_blank" rel="nofollow noopener">blackhat.com/asia-26/arsena… 🔗 Kaggle Competition: kaggle.com/competitions/o…

DeFi and LLM tech are moving faster than ever, security must keep pace. We’re bringing #Bastet to #BHASIA to showcase our latest research. ⚡️Join the conversation⚡️ - Deep dive into the Bastet Framework - Kaggle competition experience exchange - Future visions for Bastet AI-agents 📍 Black Hat Asia Arsenal 📅 April 23 (4:00pm) | April 24 (11:30am) Let's secure Web3 together! 🐈‍⬛ 🐈‍⬛🐈‍⬛ #BHA2026 #OneSavie #DeFiHackLabs #Bastet

Here's the setup: 70 real exploits on Ethereum and BSC across 8 vulnerability categories. Ground-truth labels adapted from @DefiHackLabs' DeFi Security Breach RCA dataset. Huge credit to them for putting up the repo: github.com/SunWeb3Sec/DeF… I tested 7 configurations from @OpenAI (GPT-5, GPT-5.3-Codex, GPT-5.4 High/xHigh) and @AnthropicAI (Sonnet 4.6 High, Opus 4.6 High/Max) in their native agent runtimes. 490 runs total, each graded by two independent LLM judges.

Introducing TxAnalyzer, an open-source AI-powered blockchain attack transaction analyzer Pulls traces, opcodes, and contract artifacts with one command, then automatically identifies root causes of exploits via AI Agents. Runs as a Skill — plug into Claude Code / Cursor github.com/BradMoonUESTC/…

#Bastet @Kaggle Competition: Quick Reminders 📝 To keep things smooth, here’s what you need to know: 1. Leverage the Tag Definitions 2. Fewer False Positives Lead to Higher Scores 3. Review Prize Rules & Complete Pre-Registration

Honored to be recognized as one of the first 40 ETHSecurity Badge holders by @thedaofund More to build. 🙏 #EthereumSecurity #Web3Security

@1nf0s3cpt ✋

🚀 Building an exploit-skill with LLMs I let an LLM run for 3 days. 302 rounds later: ✅ 5,771 exploit techniques & patterns across 84 vulnerability types ✅ Obsidian KB integrated Turns out: Using Obsidian as a lightweight KB works insanely well for iterative learning loops.

👉Competition Link: kaggle.com/competitions/o… #Web3Security #Kaggle #SmartContract #LLM #Bastet

#Bastet Competition: LLM Identification of Smart Contract Vulnerabilities is officially LIVE on @kaggle! Step into the arena and prove your skills! 😼😼 ⏳ Duration: April 1 – July 1, 2026 This challenge is proudly supported by: 🔹 @EF_ESP 🔹 @GCCofCommons 🔹 IEEE CIS Singapore Chapter 🔹 @IEEEBlockchain

AiSec (AI Security Initiative) A new initiative by @DeFiHackLabs , advancing LLM security (LLMSec) and beyond. AiSec — securing AI agents and autonomous systems, advancing LLM-driven offense and defense. We use LLMs to scale security — finding more vulnerabilities, reducing false positives, and building stronger defenses for AI-powered systems and workflows. Focus areas: ✅Smart contract security ✅AI agent & autonomous system security ✅LLM vulnerability research & detection We’re starting to share our initial work in open-source LLM security — with more tools and research coming soon. We’re building this in the open — contributors are welcome to join us.

🔥Just open-sourced: llm-sast-scanner skill. Supports 34 vulnerability categories. Your AI coding agent already knows how to find vulnerabilities. It just doesn’t know where to look, what to trace, or when to stop flagging false positives. So I gave it a structured SAST workflow: source mapping, taint tracing, and a Judge verification step. Benchmarks: F1 = 0.991 (Claude Opus 4.6) F1 = 0.954 (GPT-5.4) 👇Contributions are welcome to help improve detection rates! #Claude #Codex #Skills #SAST

🔥 26 → 118 cross-chain protocol decoders in 3 days. Powered by self-improving LLM loops + calldata self-decoding. Still buggy. Humans still in the loop.🤣

⏳ 1 Week Countdown for Bastet Kaggle Competition ⏳ Six days remain before the starting bell. If your AI agent’s output does not follow the competition format, consider using the tag definitions created by @onesavielab as a reference and building an output classifier for submission. Let the public leaderboard speak for itself and decide who claims victory.

If you are auditing smart contracts using AI LLMs, then you are highly suited to read this. 👇The insights from @xy9301

⏳ 2-week Countdown to the #Bastet @kaggle Competition ⏳ The dataset features real-world smart contract repositories. Each one is a complete auditing target containing multiple contracts, libraries, and supporting files. Two weeks left to refine your vulnerability analysis strategy and get your AI agent ready for the challenge.