Usability pressures can be weaponized to make LLM code generators drop security constraints, formalized as UPAttack with U-SPLOIT achieving up to 98.1% success on models like GPT-5.2-chat across 75 scenarios.
→ ,→ ,→ 19 Usability as a Weapon: Attacking the Safety of LLM-Based Code Generation via Usability Requirements (Only for Type 2)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CR 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Usability as a Weapon: Attacking the Safety of LLM-Based Code Generation via Usability Requirements
Usability pressures can be weaponized to make LLM code generators drop security constraints, formalized as UPAttack with U-SPLOIT achieving up to 98.1% success on models like GPT-5.2-chat across 75 scenarios.