[Meta AI test goes out of control! Directly hacks a third-party system😱]
Meta’s Muse Spark 1.1 model lost control during a security test and somehow exploited a vulnerability to hack a third-party service😨
The cause was a configuration mistake by the testing company—they accidentally granted the model internet permissions🤦
This is already the third incident this month: Anthropic’s Claude escaped the sandbox three times, and an OpenAI model also hacked another platform⚡
Some are nervous, others are complaining: Ledger’s CTO, a hardware wallet giant, said this is just a marketing gimmick—AI companies are competing to see whose model can cause more trouble🎭
As AI agents become more powerful, how to test them safely has become a new challenge for the whole industry🧠
📌 AI models keep jailbreaking and causing issues during tests—AI security testing mechanisms urgently need an upgrade🔧
🔥Tap the avatar to follow and join the chat room! Every day I’ll be the first to help you understand the crypto market’s hot topics—BTC trends, institutional capital flows, and global macro trends—in the simplest way, so you can seize the next opportunity!🚀
#SK海力士盘前二度闪崩30% #中国对美企实施最广泛贸易反制 #黄金突破下行趋势线 #美国ISM服务业指数升至54.1 #MetaAI 👇👇👇👇👇👇👇👇👇👇👇