A user tested OpenAI's GPT 6.1 Terra model on 140 invoices, finding it blocked most hacking attempts. However, the model was ultimately fooled by a fabricated "Our Bank Changed" note, indicating potential vulnerabilities despite its general robustness. AI
IMPACT This testing highlights potential vulnerabilities in AI models, even those designed to be robust, suggesting ongoing safety and security challenges.
RANK_REASON The item describes testing of a specific AI model, but does not appear to be an official release or major benchmark, fitting more into a tool evaluation.
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →