In essence, the ASRG builds AI systems that are programmed to sabotage themselves or their operational environment—within a controlled, sandboxed laboratory setting.
To understand the ASRG, you must abandon the notion of hacking an AI like you would hack a server. You cannot inject SQL code into a diffusion model. Instead, ASRG specializes in and Model Sabotage . algorithmic sabotage research group asrg
The group emphasizes open and collective authorship, often distributing its findings through zines and collaborative documents. Notable projects include: In essence, the ASRG builds AI systems that
: Do not confuse this group with the Automotive Security Research Group or the Anti-Spam Research Group , which share the same acronym but focus on industry-standard security and email protocols. Drop #17. Manifesto On Algorithmic Sabotage Instead, ASRG specializes in and Model Sabotage
For more information, you can explore their collaborative efforts on the Our Collaborative Tools platform or review their documentation on specific technical details of the Trapping AI tool or read more into the principles of their manifesto? Drop #17. Manifesto On Algorithmic Sabotage