{"version":1,"type":"story","url":"https://digestai.news/story/m-3-framework-improves-jailbreaking-success-rates-by-up-to-53-on-six-m","json":"https://digestai.news/story/m-3-framework-improves-jailbreaking-success-rates-by-up-to-53-on-six-m.json","markdown":"https://digestai.news/story/m-3-framework-improves-jailbreaking-success-rates-by-up-to-53-on-six-m.md","slug":"m-3-framework-improves-jailbreaking-success-rates-by-up-to-53-on-six-m","headline":"$M^3$ framework improves jailbreaking success rates by up to 53% on six models","summary":"$M^3$ is a new automated red-teaming framework that models multi-turn jailbreaking as a Markov Decision Process (MDP). The system uses an attacker LLM to interact with a target LLM while maintaining a compact internal state managed by a separate state LLM. This approach avoids the heuristics and assumptions of prior methods, allowing it to operate in fully black-box settings.\n\nResearchers evaluated $M^3$ on two benchmarks and six target models. It outperformed existing multi-turn red-teaming methods, achieving up to a 53% higher Attack Success Rate (ASR) at five turns. Even with a lightweight attacker, $M^3$ still improved ASR by 26% at one turn and 34% at five turns compared to baselines using stronger attackers. The framework also escalates risk more steeply over turns and delivers more consistent performance across risk categories, with a 4.82 improvement in the Gini index.","keyPoints":["$M^3$ models jailbreaking as a finite-horizon Markov Decision Process (MDP) for automated red-teaming","Outperforms existing methods by up to 53% in Attack Success Rate (ASR) on six target models","Works with lightweight attackers while maintaining gains of 26-34% in ASR at one to five turns"],"whyItMatters":"A more effective jailbreaking framework could push AI safety research toward scalable, automated testing methods. It may also reveal new vulnerabilities in deployed models, prompting labs to rethink their defenses.","category":{"slug":"research","name":"Research","url":"https://digestai.news/category/research"},"entities":{"companies":[],"models":["$M^3$","attacker LLM","target LLM","state LLM"],"people":[]},"firstPublishedAt":"2026-10-26T04:00:00Z","updatedAt":"2026-10-26T04:00:00Z","sourceCount":1,"hasPrimarySource":true,"sources":[{"outlet":"arXiv cs.AI","title":"$M^3$: A Multi-LLM System for Multi-Turn Jailbreaking as a Markov Decision Process","url":"https://arxiv.org/abs/2610.31937","publishedAt":"2026-10-26T04:00:00Z","type":"primary","primary":true,"lead":true}],"sourceNotes":null,"discussions":[],"thread":null,"cite":{"text":"Digest AI, \"$M^3$ framework improves jailbreaking success rates by up to 53% on six models\", 26 October 2026, https://digestai.news/story/m-3-framework-improves-jailbreaking-success-rates-by-up-to-53-on-six-m","publisher":"Digest AI","title":"$M^3$ framework improves jailbreaking success rates by up to 53% on six models","datePublished":"2026-10-26T04:00:00Z","url":"https://digestai.news/story/m-3-framework-improves-jailbreaking-success-rates-by-up-to-53-on-six-m"},"generatedBy":"Written by Digest AI's editorial model from the linked sources; the sources are the record.","license":"Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse"}