Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
rubises
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
4 ms
·
1.
▲
by
rubises
6mo ago
The harder problem isn't finding vulnerabilities — it's preventing AI from violating constraints in the first place. Prompt-level safety is probabilistic. Filesystem-level constraints (mkdir 禁/behavior) are deterministic.