Preempt AI v2 Launches With 99.65% Attack Detection Accuracy
Key Takeaways
- ▸Model-agnostic architecture allows single API integration with any AI provider regardless of underlying model
- ▸Trained on 589,000+ real-world attack samples with 99.65% accuracy in threat detection
- ▸V2 adds OCR-based image injection detection and enhanced multi-language support (12+ languages)
Summary
Preempt AI has unveiled version 2 of its security platform for AI applications, offering enterprise-grade protection against prompt injection attacks, jailbreaks, and data leaks. The updated platform uses machine learning models trained on 589,000+ real-world attack samples to detect threats with 99.65% accuracy and works model-agnostically with any AI provider through a single API call. This approach allows organizations to add security layers without being locked into specific AI platforms.
Preempt AI v2 introduces several new capabilities, including enhanced prompt injection defense, jailbreak prevention, automatic sensitive data detection and encryption (SSNs, credit cards, emails), and multi-language threat detection across 12+ languages. The v2 release adds OCR-powered detection to identify prompt injections hidden in images, addressing visual attack vectors. The platform emphasizes data protection while preserving privacy through encryption, and includes real-time security metrics for enterprise monitoring. A browser extension is also available to provide protection across any AI platform.
- Real-time security metrics and automated sensitive data encryption preserve both security and privacy
