Pangea, a key provider of security guardrails, has introduced AI Guard and Prompt Guard to enhance AI security, mitigating risks such as prompt injection and exposure of sensitive data. These additions, alongside AI Access Control and AI Visibility, establish one of the most robust protection suites for AI applications available today.
"As companies race to build and deploy AI apps via RAG and agentic frameworks, integrating LLMs with users and sensitive data introduces substantial security risks," said Oliver Friedrichs, CEO and Founder of Pangea. "New attacks surface daily, requiring countermeasures to be rolled out equally fast. As a proven and trusted partner in the cybersecurity industry, Pangea constantly identifies and responds to new generative AI threats, before they can cause harm."
"I've seen firsthand how vulnerabilities in computer systems can lead to damaging real-world impacts if left unchecked. AI's potential for autonomous action could amplify these consequences," said Kevin Mandia, Founder of Mandiant, and Strategic Partner at Ballistic Ventures. "Pangea's security guardrails draw from decades of cybersecurity expertise to deliver essential defenses for organizations building AI software."
For over three years, significant investments have been made in developing a secure, scalable, and composable platform integrating 21 security services. The latest focus is on addressing AI security challenges, including prompt injection prevention and compliance-driven data redaction. These capabilities protect enterprises embedding AI deeper into products and workflows.
AI Security Services include:
- Prompt Guard - Protects generative AI applications from prompt injection and jailbreak attempts that could expose sensitive data or manipulate AI outputs. Utilizing an ensemble of analyzers, this service achieves 99% accuracy and an F1 score of 95.2.
- AI Guard - Expands on Prompt Guard by incorporating 13 additional detectors to block unwanted content before it reaches AI models and prevent unintentional data leaks. Features include:
- Detection and masking of up to 50 types of confidential information and personally identifiable data.
- Threat intelligence-powered filtering of malicious URLs, domain names, and IP addresses with support from Crowdstrike, DomainTools, and Team Cymru.
- Filtering prompts and content by language, code, self-harm, violence, sentiment, topic, profanity, competitive language, gibberish, and more.
- AuthN & AuthZ - Fine-grained access control for Retrieval Augmented Generation (RAG), ensuring only authorized personnel and systems can interact with sensitive data. Chunk-level authorization for RAG is supported, along with an open-source multipass library for permission validation.
- Secure Audit Log - Provides detailed visibility into AI system behavior by logging prompts, model usage, versioning, document sources via RAG, and application outputs. Agentic applications also benefit from tool usage tracking and action monitoring.
To demonstrate AI security principles in action, Pangea unveils the $10,000 AI Escape Room Challenge, an interactive cybersecurity competition. Participants will engage with an AI-powered escape room, testing skills in prompt engineering and evasion techniques to uncover hidden passcodes. By experimenting with AI vulnerabilities in a controlled environment, security professionals gain critical insights into evolving threats and mitigation strategies.
With AI adoption accelerating across industries, emerging risks such as prompt injection, alignment faking, and unauthorized data exposure present new security challenges. Pangea remains dedicated to pioneering robust AI security solutions, ensuring enterprises can develop and deploy AI applications with confidence. As AI technology evolves, continuous advancements in security frameworks will be necessary to safeguard innovation and operational integrity. Explore Pangea’s AI Security Services and participate in the AI Escape Room Challenge to experience cutting-edge enterprise AI protection firsthand.
Image credit: Pangea