tszzl: AI cyber offense stays dominant as every new model opens fresh attack surfaces
tszzl · x · 2026-10-22
- tszzl 引用一条关于网络安全攻防不对称的长文并表态认同：公司被攻陷的状态空间被低估，无论多少防御投入都难以根本扭转攻防不对称，且每个新模型都在开辟新的漏洞空间，短期内看不到饱和。
- 随着疯狂 vibe coding 和 agent 攻击面扩大，情况只会更糟；即便漏洞发现与修补饱和，攻击者也会转向社工，用极有说服力的模型诱骗用户点击、授权或犯错。除非削减软件规模，防御不会成为优势方。
- 补充讨论：有人指出美国政府曾威胁 Anthropic 放缓其 Mythos 网络攻击能力的对外发布，这某种程度也拖慢了防御侧进展。
More from AGI Musings
- After 9/9: what's next for open-source model security, from formal verification to red teams — saranormous · 2026-10-22
- Insider View: Why Chinese Hospitals Are Adopting AI Far Faster Than the West — 96Stats · 2026-10-22
- Researchers urge focus on cyber risk, not nanobot doomsday, as AI's real danger — nlarusstone · 2026-10-22
- AI philosopher laments being mistaken for an AI ethicist — moonsandhues · 2026-10-22
- Blogger argues we're 1/3 of the way to recursive self-improvement — imjustnewatai · 2026-10-22
- Humans Are Born With an Innate World Model Shaped by Evolutionary Priors — yunta_tsai · 2026-10-22