Gemini 4 Carbon Achieves 0% Attack Rate on Gray Swan IPI Benchmark

• •Gemini 4 Carbon scored a perfect 0% on the Gray Swan IPI benchmark.
• •The model neutralized all 15 attack attempts during testing.
• •It is now available to select trusted cyber defenders.

Google DeepMind's Gemini 4 Carbon has achieved a perfect 0% attack success rate on the Gray Swan Indirect Prompt Injection (IPI) benchmark, demonstrating its resilience against one of the most dangerous threat vectors for AI models. The benchmark tests the model's ability to withstand indirect prompt injection attacks, where malicious instructions are hidden within external data. Gemini 4 Carbon was tested against 15 attack attempts and successfully neutralized all of them, marking it as the most secure large language model in the industry. In comparison, its predecessor, Gemini 4 Argon, had a 0.7% attack success rate, while competitors like OpenAI's GPT-6 Astra faced a 8.5% failure rate. This achievement is as enterprises increasingly deploy AI models with access to sensitive data and systems. The model is currently available to a select group of trusted cyber defenders under Google's Fairwind program.