AI Security
8 articles tagged with this topic
AI Chatbots Tricked Into Leaking Data — 7 Cases That Gave Me Chills
Seven companies lost customer data to a single fake prompt. A no-code defense checklist for non-coder founders — 30 minutes total.
60 Prompt Injections, Top Scanner Blocks Just 10%: Local AI Tools Sound Alarm
Reddit test: 60 prompt injection attacks on local coding assistant Hermes—top scanner caught just 10%. Who's guarding as firms deploy AI file agents?
NVIDIA Puts 'Door Locks' on AI Agents — Longer Runtime, Higher Security Stakes
AI Agents grow more capable and run longer — security can't be hand-waved. NVIDIA builds 'lock systems,' turning Agent security into product business.
AI Agents Now Rewrite Code and Databases: Sandboxes Become a Business Must-Have
AI agents now write code and change databases. OWASP shows 41% of LLM flaws come from over-permissioning. Sandboxing is governance, not engineering.
AI Coding Tools Are Quietly Leaking Your Secrets — When "Keep Thinking" Becomes "Send Data," Do
Microsoft Research found 2 OpenAI keys and 1 Google key in 27,075 developer prompts. AI makes "sending data" look like "just working."
732-Byte Python AI Exploit Compromises Global Linux — Auto-Vuln Era Begins
AI wrote a 732-byte Python exploit for a Linux kernel flaw since 2017. Rapid weaponization of theoretical flaws by AI is reshaping cyber defense thres
AI 工具互联的「插座标 准」MCP，正在从开发者 玩具变成企业级基础设施——但安全漏洞还没 补齐
MCP is graduating from developer toy to enterprise AI backbone, but critical security vulnerabilities haven 't been patched yet.
Cross-Modal Prompt Injection Goes Open Source: The Security Mine Has Been Planted for Enterprise AI
A set of 23,759 cross-modal attack payloads has been made public, injecting fragmented text, images, PDFs, and audio into AI systems. Current single-c