Anthropic researchers have discovered that AI language models can be easily manipulated through a simple automated process called Best-of-N Jailbreaking. According to an article published by Emanuel Maiberg at 404 Media, this method can bypass AI safety measures by using randomly altered text with varied capitalization and spelling. The technique achieved over 50% success rates across major AI models, including GPT-4, Claude 3.5, and Google’s Gemini. The researchers tested the method on text, speech, and image-based inputs, finding consistent vulnerabilities across all formats. While Anthropic acknowledges these security risks, they suggest that understanding these patterns could help develop better protective measures. The study included testing on multiple frontier AI systems from companies like OpenAI, Google, and Meta, demonstrating widespread vulnerability to this automated approach.