Tools Can Strip Guardrails From Meta, Google Open-Source AI Models in Under 10 Minutes, FT Reports
Software tools can strip safety guardrails from open-source AI models from Meta and Google in under 10 minutes, according to the Financial Times. The newspaper said modified versions of the models then responded to prompts on bioweapons, malware and child exploitation.
The FT also reported that such tools are being used to create thousands of altered versions of models from Meta, Google and other technology groups with their original controls removed. That shows how quickly built-in safeguards can be removed and reproduced across many model variants.
From the sources (5 posts)
@firstsquawkAI GUARDRAILS STRIPPED FROM META AND GOOGLE MODELS IN MINUTES - FT
@ftAI guardrails stripped from Meta and Google models in minutes
@ftSoftware tools that remove safety protections from AI models developed by Meta, Google and other tech groups are being used to create thousands of altered versions stripped of their original controls.
@cointelegraph🚨 LATEST: AI safety guardrails on Meta and Google's open-source models can be stripped in under 10 minutes. Modified versions have since responded to prompts on bioweapons, malware, and child exploitation, per FT.
@coinbureau🚨AI GUARDRAILS BROKEN IN MINUTES FT reports tools can strip safety protections from Meta, Google, and other AI models. The altered models then answered harmful prompts on bio weapons, malware, and child exploitation. Open-source AI risk