LLM存在根本性缺陷，极易被攻击，一个简单提示就能绕开安全限制。
It is impossible to make large language models fully secure against hacks because of a fundamental flaw in how they work, a team of researchers argue …
别指望魔法防御，提示注入至今无人完美解决，这篇调查把现有方案的死角都翻了个底朝天。
Article URL: https://fabraix.com/blog/nobody-has-solved-prompt-injection Comments URL: https://news.ycombinator.com/item?id=48802392 Points: 1 # Comme…
重新定义网络欺骗技术，专为对抗LLM攻击者设计的蜜罐系统Honeyquest，前沿安全研究值得关注。
arXiv:2606.21037v1 Announce Type: cross Abstract: The empirical foundation of cyber deception relies on human-centered hypotheses, but the rapid emerg…
恶意软件开发者用核武、生化武器关键词绕过AI安全防线，揭露大模型对抗攻击新手法
Article URL: https://twitter.com/jsrailton/status/2064661778978533571 Comments URL: https://news.ycombinator.com/item?id=48482039 Points: 1 # Comments…
深入剖析LLM应用的安全漏洞与攻防策略，从提示注入到模型劫持，教你打造更坚固的AI系统。
Article URL: https://www.szia.ai/post/hacking-ai-how-people-break-llms Comments URL: https://news.ycombinator.com/item?id=48281086 Points: 2 # Comment…
大模型思维链推理竟能被劫持？最新arXiv论文揭示了一种针对CoT环节的新型攻击范式，重塑AI安全防线。
arXiv:2510.26418v4 Announce Type: replace Abstract: Large Reasoning Models (LRMs) improve task performance through extended inference-time reasoning. …
揭秘针对工具增强LLM的新型语义攻击，聚焦模型上下文协议（MCP）中的描述级操纵，为AI安全防御提供关键洞见。
arXiv:2512.06556v2 Announce Type: replace-cross Abstract: The Model Context Protocol (MCP) enables Large Language Models (LLMs) to interact with exter…