AI Governance & Security Vendor Wiki

3 min read

Cisco AI Defense

Primary category: ai-runtime-security. Built on the former robust-intelligence, now owned by cisco.

One-liner — Cisco’s “security for AI” product: it validates and red-teams models before deployment and enforces runtime guardrails (prompt-injection, jailbreak, data-leak, toxic-content) on AI apps in production.

What it does — Two halves. (1) Validation / algorithmic red teaming — automatically probes models and AI applications for vulnerabilities (jailbreaks, prompt injection, data poisoning, unsafe outputs) before they ship, the capability that came from robust-intelligence. (2) Runtime protection — an inline guardrail layer that inspects prompts and responses to block attacks and policy violations in production. Cisco positions it as model-agnostic protection that plugs into its network/SASE fabric so the same controls apply wherever AI traffic flows.

Where it sits in the stack — Squarely ai-runtime-security (the AI firewall) with a strong ai-red-teaming component. Trifecta role: breaks the untrusted-input leg (prompt-injection / jailbreak defense), and helps on sensitive-data and egress by inspecting model I/O. Lives at the model/prompt layer, fed by and feeding Cisco’s network and SOC (splunk) zones.

Deployment & architecture — Delivered as a Cisco cloud service with API/SDK enforcement points and inline inspection that can ride Cisco Secure Access / network egress. Integrates with the broader Cisco Security Cloud and routes detections to splunk. Model-agnostic across major LLM providers.

Positioning & differentiators — Among the earliest “AI firewall” entrants by pedigree — Robust Intelligence shipped one of the first AI firewalls and validation platforms out of Harvard ML research. Cisco’s edge over standalone rivals is distribution: bundling AI Defense with network, identity (Duo), and SIEM that enterprises already own. Nearest neighbors: prisma-airs, hiddenlayer, witnessai, pillar-security, lakera, calypsoai.

Ownership, funding & M&A

A Cisco product, not an independent vendor. Robust Intelligence — founded 2019 in San Francisco by Yaron Singer (CEO), Kojin Oshiba, Eric Balkanski and Alexander Rilee — was acquired by Cisco (intent 2024-08-26, closed ~2024-09-24) and its technology became Cisco AI Defense (and Cisco Foundation AI). Confirmed against Cisco’s announcement and the “Robust Intelligence is now part of Cisco” page (high confidence). See robust-intelligence (alias page) for the legacy entity.

CTO / hedge-fund lens

Day-1 if you are exposing internal or customer-facing LLM apps and want a single product that both pre-deployment red-teams and runtime-guards them. Strongest fit for a fund already standardized on Cisco/Splunk — the integration story is the reason to pick it over a pure-play. Red-teaming/validation output is useful SR 11-7 / model-risk evidence (documented testing of model behavior), though Cisco AI Defense is a security tool, not a governance system of record. If you have no Cisco footprint, weigh best-of-breed pure-plays on detection quality.

Competitors / alternatives

prisma-airs, hiddenlayer, witnessai, pillar-security, lakera, calypsoai, enkrypt-ai, trojai, splxai.

Open questions / to verify

  • Current pricing/packaging and whether validation and runtime are sold separately.
  • Depth of model coverage and how guardrail detection benchmarks vs pure-plays.

Sources

History

  • [2026-06-28] Stub created from seed registry.
  • [2026-06-28] Researched; confirmed Cisco AI Defense is built on Robust Intelligence (founded 2019 SF; Cisco intent 2024-08-26, closed ~2024-09-24). Set ownership=subsidiary (Cisco), confidence high. Merged robust-intelligence into this page as the canonical profile; that slug is now a thin alias.

Graph View

Backlinks