Vendor evidence

Knostic AgentMesh claims continuous discovery, version tracking, SHA-256 artifact identification, static-code scanning, and risk analysis for AI agent skills, Model Context Protocol (MCP) servers, integrated development environment (IDE) extensions, Claude Code plugins, and commands.

Knostic Kirin claims runtime monitoring, prompt-injection blocking, unsafe-action prevention, least-privilege policy, and consistent guardrail enforcement across software as a service (SaaS), in-house, and Model Context Protocol (MCP) agents.

Knostic claims OAuth-based agent authentication, agent-role to privilege mapping, least-privilege and need-to-know boundaries, and policy enforcement across software as a service (SaaS), in-house, and Model Context Protocol (MCP) agents.

Knostic claims endpoint guardrails plus AgentMesh scanning for coding assistants, Claude and Cursor skills, Model Context Protocol (MCP) servers, integrated development environment (IDE) extensions, plugins, commands, code injection, cryptomining, reverse shells, and other agentic supply-chain threats.

Onyx AI claims a approved self-hosted enterprise AI platform with organization and group sharing, permissions, usage analytics, model controls, auditability, and governed access to enterprise knowledge and actions.

Onyx AI materials reviewed did not provide a public product claim for automated adversarial testing, repeatable attack suites, model or agent red teaming, or release-gate evaluation.

Onyx AI materials reviewed did not establish model artifact scanning, provenance, signing, dependency or Model Context Protocol (MCP) component analysis, tamper detection, or model-registry release controls.

Onyx AI claims configurable agents with controlled Actions, data access, knowledge sources, and the ability for administrators to enable or disable Model Context Protocol (MCP) and OpenAPI actions.

Onyx AI claims organization and group-based agent sharing plus configurable knowledge and action permissions for enterprise agents.

Onyx AI materials reviewed did not establish governance of coding-agent commands, developer-workstation files or networks, integrated development environment (IDE) extensions, skills, hooks, secrets, or package actions.

Cato AI Security claims discovery and inventory of managed, local, and shadow agents, centralized runtime visibility, security posture, compliance-risk remediation, and corporate policy across users, agents, applications, models, and tools.

Cato AI Security and AIM materials reviewed did not provide a public customer-facing product claim for automated adversarial testing, repeatable attack suites, model or agent red teaming, or release-gate evaluation.

Cato and AIM claim continuous AI-SPM scanning of internal models, agent configurations, Model Context Protocol (MCP) connections, training environments, misconfigurations, and vulnerabilities before production.

Cato AI Security claims runtime policy across user prompts, model outputs, tool calls, and tool messages, blocking or redacting sensitive data and indirect prompt injection before content reaches an agent or model.

Cato AI Security claims visibility into agent configurations, licenses, data and tool access, plus corporate policy enforcement and blocking of unauthorized access or unsafe actions across managed and local agents.

Cato claims endpoint discovery and runtime control for Cursor, Claude Code, and other coding agents, covering configurations, licenses, Model Context Protocol (MCP) servers, prompts, model outputs, tool calls, tool messages, credentials, PII, and malicious tool responses.

JetStream SAIG claims AI estate discovery, approved design control, accountable ownership, runtime governance, drift detection, audit-ready evidence, policy boundaries, and cost accountability across humans, agents, NHIs, models, tools, Model Context Protocol (MCP), software as a service (SaaS), and cloud.

JetStream SAIG materials reviewed did not provide a public product claim for automated adversarial testing, repeatable attack suites, model or agent red teaming, or release-gate evaluation.

JetStream claims visibility into model swaps, toolchain expansion, Model Context Protocol (MCP) servers, open-source components, configurations, credentials, permissions, connections, and runtime drift with Verified Model Context Protocol (MCP) and approved-design governance.

JetStream claims inline runtime enforcement of approved AI workflow designs across agents, tools, models, Model Context Protocol (MCP) servers, data, permissions, and identities with continuous behavioral telemetry and drift response.