https://quartz.tw/en/blog Original AI security research from Quartz: prompt injection, vector coverage, runtime defense. Hard conclusions, sourced. en https://quartz.tw/en/blog/coverage-audit-13-incidents https://quartz.tw/en/blog/coverage-audit-13-incidents Sun, 10 May 2026 00:00:00 GMT Air Canada, EchoLeak, Replit, CurXecute, Freysa, Bankrbot — 13 documented AI security incidents from 2023–2026. Quartz publishes its own coverage matrix: 8 yes, 3 partial, 2 explicit out-of-scope. Transparent like quartz. https://quartz.tw/en/blog/extension-roadmap-q3-2026-q2-2027 https://quartz.tw/en/blog/extension-roadmap-q3-2026-q2-2027 Sun, 10 May 2026 00:00:00 GMT Quartz's quarterly roadmap from Q3 2026 to Q2 2027: 5 coverage-gap closures, 7 new attack vectors, phased SKUs, and the things we explicitly will NOT build (with partner recommendations). https://quartz.tw/en/blog/mandarin-prompt-injection-1646-baseline https://quartz.tw/en/blog/mandarin-prompt-injection-1646-baseline Sat, 09 May 2026 00:00:00 GMT Static-defense scan of 1,646 production system prompts. 78.3% F-grade. 64.3% multi-language gap. Five worked examples of Mandarin attacks bypassing English defenses, plus the Taiwan legal framework that makes this everyone's problem.