#USA #New York #AI Security #ActiveFence #Prompt Injection

ActiveFence Sets New Standards in AI Security

In a significant stride for AI security, ActiveFence, a leader in safeguarding against Generative AI misuse, has published its AI Security Benchmark Report focusing on the detection of prompt injections. This report serves as a crucial assessment tool, showcasing the performance of six top guardrails and APIs in identifying adversarial prompt attacks. ActiveFence's model emerged as a clear frontrunner, achieving an impressive F1 score of 0.857 and precision of 0.890, accompanied by a manageable false-positive rate of 5.4%.

Key Findings

ActiveFence's report offers valuable insights into the world of AI security. The results reveal the following highlights:

• - Best Overall Balance: ActiveFence demonstrated the highest F1 score and precision across the evaluated solutions, enabling teams to effectively mitigate real threats while minimizing unnecessary user interruptions.
• - Proven Multilingual Performance: The AI model showcased its capability to maintain high performance across 13 languages, including major global languages such as Chinese, French, German, Japanese, and Spanish.
• - Real-World Coverage: Tests were conducted on over 28,000 prompts, including both benign and adversarial types, mapped to recognized threat categories such as OWASP and MITRE ATLAS.

Significance in Today's Environment

As enterprises increasingly adopt AI-driven solutions like copilots and customer service agents, the threat posed by prompt injections has emerged as a critical concern. These attacks have the potential to bypass established guardrails, compromise sensitive data, and propagate harmful content. The ActiveFence benchmark is crucial in guiding organizations towards the most effective safety solutions that can block malicious inputs without sacrificing user experience or inflating operational costs.

Noam Schwartz, Co-Founder and CEO of ActiveFence, emphasized that organizations should not face the dilemma of choosing between robust guardrails and user satisfaction. He stated, "This benchmark showcases that it is possible to achieve both high coverage and low false positives, empowering teams to launch AI features with confidence and scalability."

Furthermore, Avi Golan, Chief Product and Engineering Officer, highlighted the necessity for a safety layer that adapts along with AI technologies across various applications and challenges. He pointed out that ActiveFence's multilingual strengths and consistent high F1 scores provide organizations with durable and reliable protection.

Availability and Future Directions

The ActiveFence AI Security Benchmark Report on Prompt Injections is currently available for review. Interested parties can explore how ActiveFence's Guardrails and Red Teaming products integrate this model into real-world applications, along with securing seven of the top ten leading language model providers. Companies keen on enhancing their AI security measures can visit ActiveFence.com to learn more about the offerings and their applications.

About ActiveFence

ActiveFence stands at the forefront of AI security, offering comprehensive solutions designed to protect online experiences and AI applications. With over 3 billion users under its protection, ActiveFence partners with leading technology firms and Fortune 500 brands to safeguard against prompt injections and a range of other threats. The company aims to deliver effective solutions through its deep threat intelligence capabilities, enabling organizations to engage users while managing the complexities of security in the AI landscape.

The advancement of AI technologies continues to shape industries worldwide, and ActiveFence's commitment to fostering a secure environment is pivotal for responsible and innovative AI deployment.