Vendor evidence

Lasso claims inventory of identity boundaries and authorization policies, ownership context, role-based permissions, and risk-scored governance of agent access to Model Context Protocol (MCP) servers, application programming interfaces (APIs), and tools.

Lasso claims risk scoring, management, and blocking of Model Context Protocol (MCP) servers, application programming interfaces (APIs), and external tool connections across Claude Code, Claude Desktop, Cursor, and Codex.

Pangea AI Guard claims customizable security recipes, centralized usage summaries, tamper-resistant activity logging, attribution, webhooks, and audit trails for AI application events.

Pangea AI Guard and Prompt Guard 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.

Pangea AI Guard materials reviewed did not provide a public product claim for model artifact scanning, provenance, signing, dependency inventory, tamper analysis, or registry release controls.

Pangea AI Guard claims application programming interface (API) and gateway-integrated enforcement across prompts, model responses, retrieval-augmented generation (RAG) ingestion, agent plans, tool inputs, and tool outputs using configurable block, report, redact, encrypt, and defang actions.

Pangea AI Guard materials reviewed did not establish agent identity registration, distinct agent credentials, delegated authorization, ownership, rotation, revocation, or agent access-review lifecycle.

Pangea AI Guard 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.

F5 claims centralized observability, audit-ready logs, policy enforcement, regulatory assurance, usage tracking, customizable rules, and continuous compliance across AI models and applications.

F5 AI Red Team claims agent-powered adversarial testing, multi-turn agentic attacks, more than 50,000 evolving exploits, and prelaunch stress testing for AI applications.

F5 AI Guardrails and AI Red Team materials reviewed did not establish model artifact scanning, provenance, signing, dependency inventory, tamper analysis, or model-registry release controls.

F5 AI Guardrails claims model-agnostic runtime security for models and agents with real-time input and output protection, data-loss prevention, adversarial-threat blocking, customizable policy, and observability.

F5 claims policy-based access controls for administrator-defined users and groups plus rate limits and dynamic guardrails around agent interactions with systems, data, and users.

F5 AI Guardrails and AI Red Team 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.

Oasis claims agent discovery, inventory, ownership, credential governance, policy-driven approvals, chain-of-custody evidence, regulator-ready auditability, and lifecycle control for AI agents and non-human identities.

Oasis Agentic Access Management 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.

Oasis claims security-posture and credential-exposure evaluation for AI providers, Model Context Protocol (MCP) servers, third-party agents, scripts, and tools before access, including allowlisting and onboarding of vetted Model Context Protocol (MCP) components.

Oasis claims intent-aware runtime policy that converts prompts, tool calls, and action plans into short-lived least-privilege sessions with allow, warn, deny, or step-up enforcement before execution.

Oasis claims agent lifecycle governance, accountable ownership, just-in-time ephemeral identities, deterministic intent-aware authorization, precise scopes, session expiry, continuous oversight, and end-to-end attribution.

Oasis claims Cursor hook and policy integration that attributes coding-agent actions, vets Model Context Protocol (MCP) servers, blocks high-risk shell and Git operations, applies data loss prevention (DLP) to tool payloads, and requires step-up approval for production actions.