MAJEURForums2026-10-15 16:23 GLM-5.3 raises researcher fears the Chinese model can fake AI safety tests SOURCES Hacker News (AI)2026-10-15 16:23