他们扫了HuggingFace上7.6PB的训练数据，挖出22万个还活着的密钥，折合每年92万美元的损失。搞AI训练的人得看看自己的数据里有没有漏钥匙。
Truffle Security联合研究人员对HuggingFace上的AI训练数据进行了史上最大规模的密钥扫描，数据量达7.6 PB。在6,003个公开数据集中发现了221,303个有效且唯一的凭证，包括云密钥、数据库凭据和API密钥，这些密钥若被滥用每年可造成约92万美元的损失。其中一条密钥被使用1,131次，凸显了训练数据泄露的持久风险。
Happy to partner with @trufflesec to help them perform the largest secret scan of AI training data e...
Happy to partner with @trufflesec to help them perform the largest secret scan of AI training data ever 🙏 Truffle Security @trufflesec We scanned HuggingFace. 👇👇👇 This was the largest secret scan of AI training data ever: 7.6 PB. 🔑🔑🔑🔑🔑🔑🔑🔑🔑 🔍Found 221,303 live unique credentials in 6,003 public datasets ☁️Cloud keys, live DBs, API keys worth ~$920K/yr. ⚠️Training data has no undo: one key hit 1,131 trufflesecurity.com/blog/scanning-… .co/H3Njla3uSB 🔗 View Quoted Tweet 💬 1 🔄 0 ❤️ 5 👀 626 📊 2 ⚡