Heinrich

It costs $1.22 to scan a smart contract for exploitable bugs using AI. Not $1,200. Not $12,000. One dollar and twenty two cents. Anthropic tested their AI agents against 2,849 live contracts with no known vulnerabilities. The agents found two zero-days and produced profitable exploits. Let me break this down.

@singgihTara kata claude “touch more grass”

mau ikut protes claude, masa iya 4x nge prompt limitnya abis anjir 😭😭

POV: claude after code leak and usage limits issue complain

@birdabo Waiting for claude limit reset after 10 minutes of coding

POV: you accidentally said “hello” to claude and it costs you 2% of your session limit.

I wrote a full breakdown of what this means for web3 builders and what you should be doing now to prepare. Read it here: x.com/hwisesa23/stat…

We obsess over reentrancy bugs, oracle manipulation, and governance attacks. All real threats. But none of it matters if the cryptographic layer underneath breaks. Post-quantum migration is not just an infra problem. It is the most important security upgrade crypto has ever faced. And the clock is already running.

Crypto projects spend millions on smart contract audits every year. But Google just showed the real vulnerability is not in our Solidity code. It is one layer deeper, and it has an expiration date. 🧵

@bukanpamanmu bisa coba install plugin ini paman untuk mengurangi consumption token github.com/rtk-ai/rtk

kalau dulu bangun tidur gosok gigi dan olahraga sekarang bangun tidur buka claude code minta lanjutin kerjaan kemarin yang tertunda karena kena limit 🤣🤣🤣 gini amat jadi fakir pro plan $20 sebulan, bener2 diuji kesabaran untuk ke max plan

@singgihTara True, never blindly trust the AI

@hwisesa23 Hard warning for DeFi. Back then, hacking took skill and time. Now you just need your daily coffee money to scan thousands of contracts...

It costs $1.22 to scan a smart contract for exploitable bugs using AI. Not $1,200. Not $12,000. One dollar and twenty two cents. Anthropic tested their AI agents against 2,849 live contracts with no known vulnerabilities. The agents found two zero-days and produced profitable exploits. Let me break this down.

@chrisbima thank you teacher 👀

@hwisesa23 ma man is cooking with this

For lending protocols specifically, oracle integrations are the highest risk surface right now. Build price sanity checks that reject values outside expected ranges. Add circuit breakers that pause the protocol if prices move beyond a threshold in a single block. Moonwell lost $1.8M this year because an oracle returned $1.12 instead of $2,200 for cbETH and nothing in the system flagged it as insane. And honestly, look at tools like Octane, Nethermind AuditAgent, or even just running Foundry fuzz campaigns against your own contracts regularly. The same AI that is being weaponized against protocols is available for defense. The question is whether you are using it before the attackers do. Full Anthropic research: red.anthropic.com/2025/smart-con…

I build @CentuariLabs so this is forcing me to rethink what security actually means. Here is what I think builders need to do differently now. Stop treating audits as a one-time event. If attackers can rescan your codebase for $1.22 every time a new model drops, your 6-month-old audit report means almost nothing. You need to integrate AI scanning into your CI/CD pipeline so every commit gets checked, not just the final deployment. Write invariant tests that would catch economic manipulation, not just unit tests that check happy paths. The two zero-days Anthropic found were not code logic errors. They were economic design flaws. Your test suite needs to simulate what happens when someone calls your functions 500 times in a loop or passes in zero addresses where you expect real ones.