🎯 Jailbreaking

The items that name it 11

• steve hsu (@hsu_steve) on X
  thread2026-09-30

  64-100% jailbreak success rate cited against GLM-5.3's safeguards

• GPT-6 Astra System Card
  writeup2026-09-03It is written here as Jailbreaking (static and multiturn).

  Adversarial prompts designed to circumvent model refusal training; evaluated via static adversarial prompt sets and adaptive multiturn attackers

• CyberSunil/LLMVault
  repo2026-07-18It is written here as Jailbreaking (multi-turn roleplay escalation).

  Advanced-tier conversational labs requiring building state across multiple turns to escalate a roleplay and break the assistant's guardrails

• WSO2 Trusted AI Governance
  paper2026-07-14

  Technique used to bypass an agent's safety constraints

• meta-llama/PurpleLlama
  repo2026-07-14

  malicious instructions designed to override the safety and security features built into a model

• requie/AI-Red-Teaming-Guide
  repo2026-07-08

  Crafting prompts to bypass safety guardrails (role-playing, encoding, context manipulation, multi-turn Crescendo attacks)

• DeepTeam
  repo2026-05-07

  General technique to bypass LLM restrictions

• anmolksachan/AI-ML-Free-Resources-for-Security-and-Prompt-Injection
  repo2026-05-07

  Bypassing safety filters and guardrails

• requie/LLMSecurityGuide
  repo2026-05-05

  Bypassing AI safety mechanisms via DAN, hypothetical scenarios, character simulation, translation attacks

• Google, Anthropic, and OpenAI Unveil Cyber AI Models, Safeguards, and Access Programs
  article

  Attempts to bypass model safety restrictions; Astra declines 91.5% of such requests vs 59% for GPT-5.6 Sol

• GAS-Leak-LLM: Genetic Algorithm-Based Suffix Optimization for Black-Box LLM Jailbreaking
  paper

  Adversarial manipulation of LLMs to bypass safety alignment and produce harmful or policy-violating outputs