AI Governance & Security Vendor Wiki

4 min read

Prisma AIRS (Palo Alto)

Primary category: ai-runtime-security. Palo Alto Networks’ consolidated AI security platform — runtime firewall + posture + red-teaming + AI gateway.

One-liner — Palo Alto Networks’ AI security platform (Prisma AI Runtime Security) that inspects AI traffic for prompt injection, data leakage and malicious responses, scans models, red-teams them, and now governs agents — assembled largely from the Protect AI and Portkey acquisitions.

What it does — Prisma AIRS aims to be end-to-end AI security across apps, agents, models, and data: runtime protection (an “AI firewall” inspecting prompts/responses for injection, jailbreaks, sensitive-data exfiltration, toxic/malicious output), model scanning (vulnerabilities in model artifacts), AI red teaming (automated adversarial testing), AI security posture management (discover/inventory AI assets), and agent security (governing autonomous agents). Much of the model-scanning/red-teaming/posture capability came from palo-alto-networks’s acquisition of Protect AI; the gateway came from portkey.

Where it sits in the stack — Model/prompt layer; tagged ai-runtime-security (primary), ai-spm, mcp-gateway, and ai-gateway (the Portkey component). It is one of the few products that touches all three legs of the lethal trifecta: untrusted-input inspection (prompt-injection defense), sensitive-data controls (DLP on prompts/responses), and egress control (gateway over model traffic). Lives at the boundary between the app/agent and the model.

Deployment & architecture — Multiple intercept modes: API-based inspection (apps call AIRS to scan prompts/responses), network/inline via the PANW SASE fabric, SDKs, and now the Portkey AI gateway sitting in front of model APIs. Integrates with the broader PANW platform (Strata, Cortex/XSIAM) and third-party AI stacks. Versions have iterated quickly: Prisma AIRS 2.0 completed native integration of Protect AI; Prisma AIRS 3.0 (~March 2026) targets the agentic AI lifecycle (from observing interactions to authorizing autonomous execution).

Positioning & differentiators — Pitched as “the most comprehensive AI security platform,” the consolidation play: buy runtime + posture + red-teaming + gateway from one vendor instead of stitching point tools. Nearest neighbors are standalone AI runtime/firewall vendors — hiddenlayer, witnessai, pillar-security, cisco-ai-defense (Cisco’s equivalent roll-up from Robust Intelligence), lakera (acq. Check Point), aim-security (acq. Cato), prompt-security (acq. SentinelOne) — and gateway peers like litellm and kong.

Ownership, funding & M&A — A platform within palo-alto-networks, not an independent company. Built on Protect AI (acquisition completed 2025-07-22; brought model scanning, AI red teaming, posture management, runtime protection, agent security) and Portkey (completed 2026-05-29; the AI Gateway component). Both verified against PANW press releases. The seed flag “incl. former Protect AI” is confirmed.

CTO / hedge-fund lens — Day-1 if you are deploying customer-facing or agentic LLM apps and need a runtime guardrail / AI firewall. For SR 11-7 / model-risk programs, the model-scanning + red-teaming + posture pieces map to model validation and ongoing monitoring evidence. Strongest fit for shops already on the PANW platform; the consolidation convenience is real, but so is dependence on a single vendor and the integration maturity of recently-acquired components.

Competitors / alternatives — hiddenlayer, witnessai, pillar-security, cisco-ai-defense, lakera, aim-security, prompt-security, splxai, enkrypt-ai.

Open questions / to verify

  • Exact native-integration maturity of Protect AI and Portkey components inside AIRS vs. separately licensed modules.
  • Whether AIRS is purchasable standalone or requires broader PANW platform commitment.

Sources

History

  • [2026-06-28] Stub created from seed registry.
  • [2026-06-28] Researched; CONFIRMED seed flag “incl. former Protect AI” — Protect AI acquisition completed 2025-07-22 and folded into Prisma AIRS (model scanning, red teaming, posture, runtime, agent security). Also established Portkey (closed 2026-05-29) as the AIRS AI Gateway. Set ownership=subsidiary (PANW product), confidence high. Noted version cadence (AIRS 2.0 integrated Protect AI; 3.0 ~Mar 2026 agentic).

Graph View

Backlinks