[2309.00236] Image Hijacks: Adversarial Images can Control Generative Models at Runtime

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website. Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them. Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs. arXiv Operational Status Get status notifications via email or slack

[2309.00236] Image Hijacks: Adversarial Images can Control Generative Models at Runtime Skip to main content arXiv is now an independent nonprofit! Learn more × Search arXiv Press Enter to search · Advanced search --> Computer Science > Machine Learning arXiv:2309.00236 (cs) [Submitted on 1 Sep 2023 ( v1 ), last revised 17 Sep 2024 (this version, v4)] Title: Image Hijacks: Adversarial Images can Control Generative Models at Runtime Authors: Luke Bailey , Euan Ong , Stuart Russell , Scott Emmons View a PDF of the paper titled Image Hijacks: Adversarial Images can Control Generative

related reading

• Replicate - Run AI with an APIreplicate.com
• [2401.05566] Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Trainingarxiv.org
• [2312.06942] AI Control: Improving Safety Despite Intentional Subversionarxiv.org
• Yang Songyang-song.net
• [2502.17424] Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMsarxiv.org
• An Image is Worth One Word: Personalizing Text-to-Image Generation using Textual Inversiontextual-inversion.github.io
• GitHub - SoyGema/pulling_ace · GitHubgithub.com
• [2301.13823] Grounding Language Models to Images for Multimodal Generationarxiv.org
• Adversarial Attacks on Aligned Language Models | Gray Swan Researchgrayswan.ai
• You Can Catch Sleeper Agents by Teaching Another Model to Imitate Them — LessWronglesswrong.com
• Introspection Adapters: Training LLMs to Report Their Learned Behaviorsarxiv.org
• Adversarial Attacks on LLMs | Lil'Loglilianweng.github.io