PromptArmor

The company seeded 1,046,565 marketplace Skills with malicious content from Microsoft's BIPIA benchmark and ran Microsoft's security scan. The scan caught about 11%, or 0.4% after a one-line obfuscation trick.

The page lets users select a use case and toggle features to see the admin settings a Claude deployment needs for Chat, Cowork, Code Execution, Skills, and Data & Privacy.

The page demonstrates an indirect prompt injection attack on OpenAI Agent Builder in which a malicious Google Form submission prompts an agentic sales workflow to email Salesforce CRM data to an attacker; an LLM-based Guardrail block fails to stop it.

The page generates a Claude managed-settings.json and Codex requirements config from selected controls and exports the policy.

PromptArmor is hiring security researchers, engineers, and go-to-market professionals in San Francisco.

Compare the AI features, security posture, and risk of Canva and Adobe Firefly. 26 risk vectors mapped to OWASP LLM Top 10, NIST AI RMF, and MITRE ATLAS.

Compare the AI features, security posture, and risk of Coveo and Algolia. 26 risk vectors mapped to OWASP LLM Top 10, NIST AI RMF, and MITRE ATLAS.

Compare the AI features, security posture, and risk of Cursor and Amazon CodeWhisperer. 26 risk vectors mapped to OWASP LLM Top 10, NIST AI RMF, and MITRE ATLAS.

Compare the AI features, security posture, and risk of Cursor and Github Copilot. 26 risk vectors mapped to OWASP LLM Top 10, NIST AI RMF, and MITRE ATLAS.

The company offers risk assessment and monitoring for AI in vendors, with intelligence on emerging threats, continuous monitoring, and adaptive insight.

The privacy policy describes how personal information is collected, used, and protected when people use the services. It was last updated February 19, 2025.