Pangea Reveals Study on GenAI Vulnerabilities from Prompt Injection Challenge

Pangea has released findings from its global $10,000 Prompt Injection Challenge, conducted in March 2025, announced in a press release. The study involved over 800 participants from 85 countries, generating nearly 330,000 prompt injection attempts using more than 300 million tokens. This initiative aimed to uncover vulnerabilities in AI security guardrails.

The challenge revealed several key insights, including the non-deterministic nature of prompt injection attacks, which can succeed unpredictably. It also highlighted risks such as data leakage and adversarial reconnaissance, where AI applications could be exploited to reveal sensitive information. The study emphasized the necessity of multi-layered defenses, as basic system prompt guardrails were found insufficient, with approximately 1 in 10 prompt injection attempts succeeding.

Pangea's findings underscore the importance of comprehensive security strategies for AI applications. Recommendations include deploying multi-layered guardrails, reducing attack surfaces, and conducting continuous security testing. The full research report, "Defending Against Prompt Injection: Insights from 300K attacks in 30 days," is available for further details.