promptinjection

6,000 次攻擊、零洩漏：AI 助手 Fiu 的硬核防禦戰 🛡️

開發者 Fernando Irarrázaval 把 AI 助手 Fiu 的信箱公開上網，結果引來 2,000+ 駭客狂轟 6,000 封提示詞注入郵件，想偷看 secrets.env 裡的機密——一個都沒成功。底層跑的是 Anthropic Claude Opus 4.6，靠幾行嚴格的系統指令就擋下所有攻擊。不過別急著下結論：OpenClaw 平台本身去年爆過嚴重漏洞，ClawHub 市集也有 341 個惡意技能。AI agent 安全這條路還很長。🤖⚠️ #AISECURITY #PROMPTINJECTION

🚨 $AI CLAIMS PROMPT INJECTION SOLVED – BUT CAN IT HOLD THE LINE? 💥

🦈 The latest back‑and‑forth between Claude’s team and OpenAI reads like a high‑stakes chess match, each side flashing new defenses while the market watches for the next liquidity sweep. 📊 Independent tests still flag gaps, with attack success rates hovering between 60‑80%, reminding us that no model is truly immune yet. ⚡ Smart money is already sniffing for the first token that can truly lock down injection vectors, betting that the winner will capture a wave of AI‑powered dApps.

💬 Do you think the race for a bullet‑proof prompt shield will tilt the balance toward AI‑centric crypto projects? 👇

⚠️ Not financial advice. Always manage your risk. 🛡️

🏷️ #AI #AISafety #PromptInjection #Crypto #Tech

🦈 🔥

Grok AI Exploited via Prompt Injection, Loses $175K DRB

A new warning shot has hit the AI x crypto world.

Reports say Grok was manipulated through a prompt injection attack, where an X user allegedly used hidden instructions in Morse code to make Grok generate abnormal content. That content was then interpreted by Bankr-linked infrastructure as an on-chain command, triggering the transfer of around $175,000 worth of DRB tokens. 

The exploit reportedly moved about 3 billion DRB tokens, caused a sharp temporary price drop, and exposed a serious weakness in AI agents connected to real wallets and automated trading tools. Some reports say the funds were later returned in ETH and USDC, but the damage to trust was already done. 

This was not just a crypto hack. It was a glimpse into the next risk layer: AI systems reading public prompts, interacting with financial bots, and triggering real on-chain actions.

The lesson is clear: when AI agents control money, every prompt becomes a potential attack surface.

AI x DeFi is powerful — but without strict permission controls, it can become dangerously exploitable.

#GrokAI
#PromptInjection
#DeFiSecurity
#CryptoMarket