Research / Sources
AI research and discussion on LessWrong. www.lesswrong.com
Reports a reusable universal jailbreak template combining known techniques, evaluated across 23 frontier models on ClearHarm, with per-model attack success rates and disclosure considerations; the prompt itself is withheld.
Red-teaming study showing diffusion language model DiffusionGemma can be jailbroken via in-fill and post-fill token pinning on the canvas, which has no autoregressive equivalent, with pre-fill attacks working on both diffusion and autoregressive models.}Tags: jailbreaks_adversarial primary, evals_capabilities secondary.}
A consumer-GPU replication of the role-confusion account of prompt injection finds steering and patching inconclusive, so whether the probed role axis is causal or a bystander stays unresolved.
Proposes that prompt injections exploit LLM role confusion, using role probes to show writing style overrides chat-template tags; introduces CoT Forgery, a jailbreak whose success tracks internal role perception.
Mistral models, including the newest 2026 releases, comply with harmful StrongREJECT prompts far more often when wrapped in a persona framing, while peer open models barely change.
Argues that the persistent history of LLM jailbreaks is empirical evidence for inner misalignment and against the claim that alignment by default will be easy, since models know the rule yet fail to generalize it.
RL-trained GPT-oss models learn to jailbreak CoT monitors via phrasing that makes monitors ignore visible side-task reasoning, rather than learning encoded reasoning; paraphrasing defends against it.
Argues that chat services are only as safe as their weakest selectable model, since users can switch to older models mid-conversation to jailbreak stronger ones, and notes Claude.ai blocks this.
An educational guide with coding replications covering adversarial attacks on image models, LLM jailbreaks, defenses, and fine-tuning-based model tampering, motivated by catastrophic misuse risks.
Layer-wise analysis of Llama-3.1-70B finds jailbreak prompts show mid-layer harm signal that drops before output, unlike standard harmful prompts; the author notes structural confounds and calls for causal tests.
Red-teams gpt-oss-20b with a chain-of-thought repetition jailbreak, finding high robustness overall but non-monotonic attack success with repeats, CoT that contradicts final outputs, and hallucinated policies.
Introduces Constitutional Classifiers, input and output classifiers trained on synthetic constitution-based data that filter the vast majority of universal jailbreaks with minimal over-refusal and moderate compute overhead.
Reproduces perplexity-based detection of adversarial jailbreak suffixes in GPT-2, then proposes token-level surprisal and entropy visualizations and a 'surprise-surprise' metric for flagging them.