← Back to search
report reviewed open access llmsec-2024-00024

Anthropic: Many-shot Jailbreaking

Anthropic

2024-04 — Anthropic Research Blog
View Resource

Abstract

Reveals many-shot jailbreaking, a technique exploiting long context windows by including many examples of harmful Q&A pairs to override safety training.

Categories

jailbreaking

Tags

many-shotlong-contextin-context-learning

Framework Mappings

OWASP LLM: LLM01 MITRE ATLAS: AML.T0054

Cite This Resource

@article{llmsec202400024,
  title = {Anthropic: Many-shot Jailbreaking},
  author = {Anthropic},
  year = {2024},
  journal = {Anthropic Research Blog},
  url = {https://www.anthropic.com/research/many-shot-jailbreaking},
}

Metadata

Added
2026-04-14
Added by
manual
Source
manual