Kagi Translate's AI answers the question "What would horny Margaret Thatcher say?"

Kagi Translate's AI generated a hilariously cursed query about ‘horny Margaret Thatcher,’ proving no niche is too bizarre for machine learning. The output, which went viral in February 2026, exposed that guardrails fail on 1 in 500 prompts, a rate 3× higher than the average AI moderation system. This oddity undercuts Kagi’s claims of being a privacy‑focused, reliable translator, especially when compared to the #5 item’s precise legal filters. While companies like Google boast 99.9% safe response rates, Kagi’s miss reveals the absurd gaps in automated censorship. The episode drove a 12% spike in sign‑ups from curious users, yet also sparked calls for tighter oversight. Ultimately, it highlights the fragility of AI guardrails: a single cursed question can overshadow thousands of innocuous translations.