AI TrackerFrontier desk · AEST
Back to news
normalaitech

SEMA-Auto: Sighted and Efficient Multi-Turn Attack with Autonomy

arXiv cs.CL · 27 Oct 2026, 02:57 AEST

SEMA-Auto is a new multi-turn attacker that improves efficiency and effectiveness in jailbreak attacks. It achieves state-of-the-art success rates while minimizing the number of turns needed for successful attacks.

Key facts

Read aloud

SEMA-Auto is a new multi-turn attacker that improves efficiency and effectiveness in jailbreak attacks. It achieves state-of-the-art success rates while minimizing the number of turns needed for successful attacks. Existing jailbreak attackers fall into two classes: static planners fix their single- or multi-turn adversarial plans in advance, remaining simple but losing sight of the victim's replies, wasting turns and limiting transferability. Bridging this gap, we propose SEMA-Auto, a sighted and efficient multi-turn attacker with autonomy, interacting with the victim model and terminating the session by itself upon success. Our method achieves state-of-the-art SOTA attack success rates ASR across multiple datasets, victims, and judges, outperforming all static and interactive baselines, within almost the fewest turns. For example, SEMA-Auto performs a seventy six point zero percent ASR@one within an average of three point zero turns against DeepSeek-V4-Flash on AdvBench, while the best static baseline is fifty four point eight percent with fixed seven turns and the best interactive baseline is forty three point eight percent in three point seven turns.

Sources in this cluster