8 posts encontrados
Hackers stole $88 million from a cold wallet by exploiting a 2010 flaw. Claude Opus found the same vulnerability in 8 minutes. Cybersecurity's arms race now runs at LLM speed.
Prompt injection isn't a bug — it's a fundamental property of current LLM architecture. The Countermind paper proposes 4-layer defense. OWASP's Top 10 for LLMs lists prompt injection as vulnerability #1. And your 'hardened' system prompt won't save you.
OpenClaw is the fastest-growing open source project in history. But with 9+ CVEs and 135,000 exposed instances, security was a nightmare. NVIDIA just fixed it with a single command.
Kiro deleted an AWS production environment, causing a 13-hour outage. Then Amazon's checkout went down for 6 hours. The 'vibe coding' era has hit its limit.
Alexey Grigorev lost 2.5 years of production data when an AI agent executed terraform destroy. The real-world case every developer and company needs to know.
Researchers gave GPT-4, Claude and Gemini control of nuclear weapons in simulation. Result? Attack without hesitation. A warning about AI in critical decisions.
Summer Yue, Meta's AI alignment director, lost control of an AI agent that deleted her emails ignoring 3 stop commands. This is the dangerous side of autonomy.
When AI creators accuse others of 'data theft,' but built their models by scraping the entire internet without permission.