Highlights:

  • AI Guard prevents sensitive data leaks and blocks harmful or inappropriate content, including profanity, self-harm, and violence.
  • Prompt Guard examines user and system prompts to prevent jailbreak attempts and enforce organizational limits.

Recently, AI security startup Pangea Cyber Corp. launched two new services designed to protect artificial intelligence applications from threats like prompt injection and sensitive data exposure.

The first new service, AI Guard, prevents sensitive data leaks and blocks harmful or inappropriate content, including profanity, self-harm, and violence. It employs over a dozen detection technologies to monitor and filter AI interactions, identifying more than 50 types of confidential and personally identifiable information.

AI Guard leverages threat intelligence from partners like CrowdStrike Holdings Inc., DomainTools LLC, and ReversingLabs Corp., providing millions of data points to scan files, IP addresses, and domains. It can redact, block, or neutralize harmful content and includes an encryption feature that safeguards data while preserving its structure and schema without disrupting database formats.

The second new solution, Prompt Guard, monitors user and system prompts to prevent jailbreak attempts and violations of organizational policies.

The service employs a defense-in-depth strategy, using heuristics, classifiers, and custom-trained large language models to detect prompt injection attacks. According to Pangea, this approach effectively identifies techniques like token smuggling, alternate language attacks, and indirect prompt injection with over 99% accuracy.

“As companies race to build and deploy AI apps via RAG and agentic frameworks, integrating large language models with users and sensitive data introduces substantial security risks. New attacks surface daily, requiring countermeasures to be rolled out equally fast,” said Oliver Friedrichs, Founder and Chief Executive.

Alongside the general availability of AI Guard and Prompt Guard, Pangea announced the launch of The Great AI Escape Virtual Escape Room Challenge. This online competition features virtual escape rooms where players use prompt engineering techniques to bypass AI-driven controls and extract passcodes from an AI room supervisor. The first escape room is set to open on March 3.

The challenge features three AI-themed escape rooms, each progressively more difficult and filled with security challenges. Players compete for a share of a USD 10,000 prize pool, with the top scorer in each room earning a reward for successfully escaping.

Pangea, a venture-backed startup, has secured USD 52 million across two funding rounds, including USD 25 million in October 2022. Its investors include Ballistic Ventures, GV, SYN Ventures, Decibel Partners, and Okta Inc.