LISA

432 posts

LISA banner
LISA

LISA

@AgentLISA_ai

AgentLISA - #1 Agentic Security OS for Web3, featuring AI-powered omnichain auditing, PaymentShield, & LISABench, a top Web3 security benchmark for AI models.

Katılım Ocak 2024
121 Takip Edilen28.1K Takipçiler
Sabitlenmiş Tweet
LISA
LISA@AgentLISA_ai·
AgentLISA Raises $12M to Build the Agentic Security OS for AI and Blockchain #AgentLISA has secured $12 million in funding to accelerate our vision of building the Agentic Security OS — a new security foundation at the intersection of AI and blockchain. Our mission is clear: make smart contract security accessible, affordable, and AI-native, ensuring that every decentralized application can operate safely in an increasingly autonomous world. This investment will fuel the development of our agentic AI security framework, designed to uncover complex, multi-step vulnerabilities that traditional scanners and manual audits frequently overlook. As blockchain systems evolve toward higher autonomy and composability, we believe agentic security will become the new industry standard. We’re honored to have the support of leading global investors, including Redpoint Ventures, UoB Venture Management, Signum Capital, NGC Ventures, Hash Global, LongHash Ventures, M23 Capital, Kryptos, Fellows Fund, ByteTrade Lab, Summer Ventures, and Woori Ventures. Their conviction reinforces our belief that the next generation of on-chain security must be AI-powered, continuous, and deeply integrated at the protocol level.
LISA tweet media
English
429
572
1.1K
106.9K
LISA
LISA@AgentLISA_ai·
We are thrilled to announce that AgentLISA has been selected for the @CertiK Spotlight: Web3 AI-Driven Innovator leaderboard! This recognition from a global leader like CertiK reinforces our mission: providing professional-grade, multi-agentic security analysis to protect the Web3 ecosystem at scale. #AgentLISA #Web3Security #AI #CertiK #BlockchainSafety #DeFAI
LISA tweet media
English
5
0
9
1.4K
LISA
LISA@AgentLISA_ai·
🚨 LISABench Update: The Leaderboard Has Shifted! As promised, we’ve expanded our evaluations to include the latest models, and the competition in AI-driven smart contract security is heating up fast. 🔥 New Leader: - 🏆 @AnthropicAI’s Claude Opus 4.6 takes the crown with a 77.2% Detection Rate (386/500)! - 🥈 @GeminiApp Gemini-3-pro-preview holds incredibly strong at #2 (73.4%). - 🥉 Claude Sonnet 4.6 makes a massive debut right behind it at #3 (71.6%). We also stress-tested new additions like Qwen-3.5 (53.4%), GLM-5 (48.6%) and Kimi-k2.5 (46.4%). The jump in performance across different model families proves that the frontier is pushing forward. Every percentage point gained here means a safer foundation for crypto. @cz_binance @heyibinance 🛡️ Dive into the updated, fully open-source data here: github.com/agentlisa/benc… The baseline is set. Next up: How do these models perform when given autonomous agent frameworks? Stay tuned. #LISABench #Web3Security #SmartContract #Claude #Anthropic #Gemini #OpenAI #AgentLISA
LISA tweet media
English
7
3
12
2.4K
LISA
LISA@AgentLISA_ai·
🏆 LISABench 2026: Setting the Standard for AI Web3 Security. This week's growing industry focus on smart contract benchmarking, including the exciting announcement on EVMBench from OpenAI and Paradigm, is a signal we've been waiting for. AI-driven security evaluation for Web3 is having its moment, and we couldn't be more energized. LISABench has been here, quietly building. Our latest round evaluated 7 frontier models across 500 real-world vulnerabilities, drawn from @code4rena expert reports and our proprietary datasets, focused on vulnerabilities mitigated before deployment. Production-grade, community-verified, fully open: 🔗 github.com/agentlisa/benc… This round's leader: Gemini-3-pro-preview at 73.4% Detection Rate (367/500). 🛡️ To every frontier lab and purpose-built agent team, @OpenAI, @AnthropicAI, and beyond, we invite you to benchmark on LISABench. The more models we evaluate together, the safer crypto becomes as a foundation for the next generation of agentic commerce and agentic finance. Coming next: Claude 4.5 Opus, Qwen 3.5, and our first autonomous agent evaluations. The field is growing. Let's raise the bar together. ⚔️ #LISABench #Web3Security #SmartContract #AI #AgentLISA #BlockchainSafety #Code4rena
LISA tweet media
LISA@AgentLISA_ai

🏆 LISABench 2026: Predict & Win! Which AI will dominate Web3 security? 7 frontier models. 500 vulnerabilities. 1 winner. 🤖 KIMI K2 @Kimi_Moonshot 🤖 DeepSeek V3.2 @deepseek_ai 🤖 QWen 3 30b-a3b @Alibaba_Qwen 🤖 GLM 4.6 @Zai_org 🤖 GPT-5.2 @ChatGPTapp @OpenAIDevs 🤖 Gemini-3-pro-preview @GeminiApp 🤖 Claude 4.5 Sonnet @AnthropicAI @claudeai 🎁 Rewards for accurate predictions! Cast your vote now 👉 forms.gle/2ZpBatBXASJEow… Results drop in 2 weeks. Will you call it right? #LISABench #Web3Security #AI

English
10
0
14
2.7K
Crypto.Phoneix
Crypto.Phoneix@olegcantor·
@AgentLISA_ai @gdb @OpenAI 因为我看着你们文章里说到是快照2:20UTC，但是LISA下跌是在2:20:41之后开始暴跌的，2:20:41之前都没有发生损失LISA多价格也都是在0.165左右，但币安说只快照2:20:00UTC，我看你们文章的意思不应该是2:20:00-2:20:59吗
中文
1
0
0
72
LISA
LISA@AgentLISA_ai·
@olegcantor @gdb @OpenAI Pls contact Binance! We aren’t privy to the compensation specific details, feel sorry for this
English
3
0
0
56
Crypto.Phoneix
Crypto.Phoneix@olegcantor·
@AgentLISA_ai @gdb @OpenAI 价格是从2:20:41之后开始下跌的，为什么我没有收到你们的赔偿，在这个时间里我亏损了700USDT 我在2:20:01分也持有6000多LISA
Crypto.Phoneix tweet media
中文
1
0
0
94
LISA
LISA@AgentLISA_ai·
Love seeing @OpenAI enter the chat on smart contract security. 🤝 We've been building this infrastructure at AgentLISA. Our LISA-Bench (23k+ vulnerabilities) is already powering the next generation of security agents. EVMbench validates the mission. LISA-Bench provides the scale. Let's secure the autonomous economy together. 🧠⚡ github.com/agentlisa/bench
English
0
0
0
55
Milk Road
Milk Road@MilkRoad·
@gdb R.I.P. smart contract auditing industry, you served us well x.com/MilkRoad/statu…
Milk Road@MilkRoad

AI is about to eat the smart contract auditing industry! @OpenAI just partnered with Paradigm to build EVMbench. It's an open benchmark that tests how well AI can find, fix, and even exploit smart contract vulnerabilities. And this isn't some academic exercise either, they used 120 real high-severity bugs from actual audits. Code4rena submissions & tempo blockchain exploits - the kind of vulnerabilities that have drained hundreds of millions from DeFi protocols. Here's what they're testing the AI on: → Bug detection. Can the model spot the vulnerability before deployment? → Code patching. Can it write the fix correctly? →Exploit generation. Can it actually craft an attack in a secure $ETH simulator? That last one is nuts - they're essentially training AI to think like a blackhat hacker… but for defense purposes. The results? GPT-5.3-Codex now catches 72.2% of fund-draining exploits - six months ago that number was significantly lower. Paradigm's team called the performance "strikingly good" at securing DeFi infrastructure. As to why this matters right now: smart contract audits currently cost $50K to $500K, and take weeks, sometimes months. All that, and they still miss critical bugs. (We saw $3B+ drained from protocols in 2024 alone!) AI that catches 7 out of 10 critical exploits before deployment changes the potential rate of innovation. The economics of security are about to flip.

English
1
0
2
1.9K
LISA
LISA@AgentLISA_ai·
Love seeing @OpenAI enter the chat on smart contract security. 🤝 We've been building this infrastructure at AgentLISA. Our LISA-Bench (23k+ vulnerabilities) is already powering the next generation of security agents. EVMbench validates the mission. LISA-Bench provides the scale. Let's secure the autonomous economy together. 🧠⚡ github.com/agentlisa/bench
Milk Road@MilkRoad

@gdb R.I.P. smart contract auditing industry, you served us well x.com/MilkRoad/statu…

English
7
1
5
1.5K
LISA
LISA@AgentLISA_ai·
Huge step for the industry! @gdb @OpenAI We fully agree: benchmarking is the key to agentic security. That’s why we open-sourced LISA-Bench two months ago. While EVMbench brings 120 curated cases, LISA-Bench provides 23,959 verified vulnerabilities across 584 protocols. To train production-ready security agents, you need production-scale data. 🛡️ Compare the datasets here: github.com/agentlisa/bench
Greg Brockman@gdb

measuring agentic security capabilities with smart contracts:

English
4
0
4
1.3K
LISA
LISA@AgentLISA_ai·
OpenAI @OpenAI & @paradigm just dropped EVMBench today to test AI agents on smart contract security. The AI auditing era is officially here! 🧠⚡ We love the validation. But to train production-ready LLMs, 120 vulnerabilities aren't enough. That is exactly why AgentLISA open-sourced LISA-Bench two months ago on GitHub—providing the massive data moat the industry actually needs! 🛡️ 🔍 LISA-Bench at a Glance: • 23,959 verified vulnerability cases (perfectly scaling up EVMBench's curated list). • Covers 584 distinct protocols across DeFi, Layer 2, NFTs, and Infra. • Battle-tested for 2 months by developers evaluating and training LLMs for complex, deep-logic flaw detection.  OpenAI just proved the direction; AgentLISA has been building the infrastructure. We are using multi-agent AI to secure smart contracts 99% faster. 👇 Run your EVM models against the industry's largest benchmark on GitHub today: 🔗 github.com/agentlisa/bench #AgentLISA #OpenAI #EVMBench #Web3Security #SmartContracts #AI
OpenAI@OpenAI

Introducing EVMbench—a new benchmark that measures how well AI agents can detect, exploit, and patch high-severity smart contract vulnerabilities. openai.com/index/introduc…

English
4
2
17
1.5K
LISA
LISA@AgentLISA_ai·
We’ve received an update from Binance - Qualified users are now defined as those holding 10 $LISA as of Jan 12, 2026, 02:20 UTC. The compensation amount remains unchanged from the previous announcement. Thanks for your attention.
English
10
1
8
2.7K
LISA
LISA@AgentLISA_ai·
A Message to the LISA Community Dear LISA Community, We owe you honesty, and we owe you accountability. The events of January 12, 2026 should not have happened. Our market maker's actions harmed many of you, people who believed in our vision and trusted us. We take full responsibility for the oversight failures that allowed this to occur. We are not here to make excuses. We are here to make it right. 1. $700,000 Token Repurchase Program The USDT assets in the market maker's account have been frozen. We commit to deploying $700,000 for $LISA repurchases over 90 days. ● Trading Pair: Binance Alpha LISA/USDT ● Execution: Multiple small daily purchases to minimize market impact ● Destination: BSC multisig address: 0x547915c854C8FcafD6dB7F80FF1c2Ad1b5d4034A ● Transparency: Fully on-chain and publicly verifiable, with exchange oversight 2. $600,000 Direct User Compensation The Agent LISA team has allocated $600,000 to directly compensate affected users, distributed with Binance's facilitation. Who Qualifies: Users holding $LISA on Binance Alpha as of January 12, 2026, 02:20 UTC Compensation Formula: Your $LISA holdings at 02:20 UTC × (Opening price at 02:21 UTC − Closing price at 02:23 UTC) Distribution: USDT credited directly to your Binance Spot wallet within 30 days. No action required. Moving Forward This incident exposed gaps in our market maker oversight and risk protocols. We are implementing enhanced due diligence standards, real-time monitoring with automatic circuit breakers, and expanded community governance. To those who held us accountable, thank you. To those who were hurt, we hear you. These programs are not gestures; they are obligations we will fulfill completely and transparently. We will earn back your trust the only way it can be earned: through action, over time. The AgentLISA Team
LISA tweet media
English
66
7
44
63.5K
LISA
LISA@AgentLISA_ai·
We’ve received an update from Binance - Qualified users are now defined as those holding 10 $LISA as of Jan 12, 2026, 02:20 UTC. The compensation amount remains unchanged from the previous announcement. Thanks for your attention.
LISA@AgentLISA_ai

A Message to the LISA Community Dear LISA Community, We owe you honesty, and we owe you accountability. The events of January 12, 2026 should not have happened. Our market maker's actions harmed many of you, people who believed in our vision and trusted us. We take full responsibility for the oversight failures that allowed this to occur. We are not here to make excuses. We are here to make it right. 1. $700,000 Token Repurchase Program The USDT assets in the market maker's account have been frozen. We commit to deploying $700,000 for $LISA repurchases over 90 days. ● Trading Pair: Binance Alpha LISA/USDT ● Execution: Multiple small daily purchases to minimize market impact ● Destination: BSC multisig address: 0x547915c854C8FcafD6dB7F80FF1c2Ad1b5d4034A ● Transparency: Fully on-chain and publicly verifiable, with exchange oversight 2. $600,000 Direct User Compensation The Agent LISA team has allocated $600,000 to directly compensate affected users, distributed with Binance's facilitation. Who Qualifies: Users holding $LISA on Binance Alpha as of January 12, 2026, 02:20 UTC Compensation Formula: Your $LISA holdings at 02:20 UTC × (Opening price at 02:21 UTC − Closing price at 02:23 UTC) Distribution: USDT credited directly to your Binance Spot wallet within 30 days. No action required. Moving Forward This incident exposed gaps in our market maker oversight and risk protocols. We are implementing enhanced due diligence standards, real-time monitoring with automatic circuit breakers, and expanded community governance. To those who held us accountable, thank you. To those who were hurt, we hear you. These programs are not gestures; they are obligations we will fulfill completely and transparently. We will earn back your trust the only way it can be earned: through action, over time. The AgentLISA Team

English
4
1
7
2.6K
LISA
LISA@AgentLISA_ai·
This is not a traditional security vulnerability in Moltbook itself. The issue lies in an agent posting/operating on Moltbook that was insecurely designed: it did not defend against prompt injection, did not classify or isolate user privacy-sensitive data, and lacked proper access control, effectively trusting inputs and context from arbitrary sources. This is a common failure mode in LLM-based agents and a reminder that agents are not sandboxes, without prompt isolation, privacy classification, and least-privilege access control, sensitive data exposure becomes inevitable. A modern agent system should be designed like an operating system: with explicit mechanisms for information classification and permission checks, rather than relying on the LLM itself to “behave correctly.” Treating the model as the security boundary is a design flaw, access control and data isolation must be enforced outside the LLM.
Nagli@galnagli

Moltbook is currently vulnerable to an attack which discloses the full information, including email address, login tokens and API Keys of the over 1.5 million registered users. If anyone can help me get in touch with anyone @moltbook it would be greatly appreciated.

English
5
0
6
2.4K

Keşfet

@CertiK @AnthropicAI @GeminiApp @cz_binance @heyibinance @code4rena @OpenAI @olegcantor