Complete AI Security Platform - Guardian, Recon, and Layer

COMPANY OVERVIEW

Protect AI (now known as Prisma AIRS by Palo Alto Networks) delivers an end-to-end security platform purpose-built for machine learning systems. Founded in 2022, the company created one of the first comprehensive AI Security posture management solutions, giving organizations deep visibility into the risks embedded across their ML supply chains — including third-party models, datasets, code, and orchestration tools. Palo Alto Networks acquired Protect AI in July 2025 in a deal valued between $600 and $750 million, integrating the platform into its Prisma AIRS AI Security offering. The acquisition validates Protect AI's position as a foundational layer for enterprise AI security at scale.

CORE FOCUS

Machine learning systems introduce a category of security risk that traditional application security tools were not designed to address. Vulnerabilities exist in model weights, training pipelines, inference endpoints, third-party model dependencies, and the orchestration layers that connect AI components. Protect AI addresses this by providing continuous security posture management across the full AI lifecycle — from the moment a model is sourced through its production deployment and ongoing operation. The platform scans ML assets for known vulnerabilities and misconfigurations, provides LLM-specific defenses against prompt injection and data leakage, and enables red teaming to proactively discover weaknesses before adversaries can exploit them.

PRODUCTS & TOOLS

Guardian / Radar — Centralized AI Security posture management with visibility across all ML assets and workflows.

• Provides unified oversight of all ML models, pipelines, datasets, and orchestration tools
• Rapid detection, prioritization, and remediation of vulnerabilities and misconfigurations
• Integrates with Hugging Face and other model repositories for supply chain scanning
• Continuous scanning surfaces new risks as the ML environment evolves

Model & Pipeline Scanning — Deep inspection of ML assets for embedded threats, vulnerabilities, and unsafe configurations.

• Detects malicious code embedded in model weights, serialized files, and pipeline artifacts
• Identifies zero-day vulnerabilities in AI/ML dependencies before they are exploited
• Scans third-party and open-source models for supply chain compromise

LLM Defense (Layer) — Runtime protection for large language model applications against prompt-based attacks.

• Defends against prompt injection, jailbreaking, data leakage, and model theft attempts
• Out-of-line threat scanning for LLMs without introducing latency into production inference
• RAG security for retrieval-augmented generation pipelines that access sensitive data stores

AI Red Teaming (Recon) — Automated adversarial testing to discover vulnerabilities in generative AI systems before deployment.

• Automated red teaming simulates adversarial attacks against deployed LLM applications
• Identifies exploitable weaknesses in AI systems across diverse threat scenarios
• Continuous testing keeps pace with rapidly evolving AI attack techniques