🧠 What until today seemed like the work of expert hackers… now an AI can do it in minutes.
Claude Opus 4.5, Claude Sonnet 4.5, and GPT-5, under Anthropic's experiments, managed to exploit vulnerabilities in smart contracts in simulated environments with a combined simulated value of ≈ USD 4.6 million in recent contracts.
And not only that: by analyzing a historical record of 405 vulnerable contracts between 2020-2025, the AI was able to exploit 207 of them (≈ 51 %), which represents more than USD 550 million in compromised “simulated funds.”
🔐 What does this finding reveal? A brutal warning for the crypto ecosystem.
• ⚠️ The automation of attacks: AI can now reproduce exploits that previously required experts — now a well-trained AI is enough.
• 🧪 The “old bugs” are not the worst: AI even discovered new vulnerabilities (“zero-day”) in contracts deployed in 2025 that had no known flaws.
• ⏳ The risk is increasing: with the falling operational costs of AI, automating these attacks becomes cheaper and more efficient — lowering the entry barrier for attackers.
• 🛡️ Defense must evolve: traditional audits are no longer enough. If the offense is with AI, the defense must also use AI + constant testing + deep reviews.
📣 🚨 “Maximum alert!
AI has just become the number one threat to smart contracts. If a crypto project does not reinforce its security right now, it could fall victim to an automated and devastating attack.
🔎 Before you used to say: “I trust the audit.” Now think: “Were they prepared for AI?”
🌐 In this new scenario, the difference between surviving or losing everything is called proactive auditing + intelligent defense.
🎯 Use this topic as an urgent call to the community.
This discovery is not just another headline, it is a structural alert for the entire DeFi / blockchain ecosystem. I recommend you use it for:
• Educate: many newcomers to crypto need to understand that the risk has changed, and quickly.
• Create awareness: for Web3 developers, entrepreneurs, investors to understand that “code + ignorance = total vulnerability.”
• Encourage good practices: constant audits, use of defensive AI, reviews before launching contracts, trust only in projects with real security.
