AI agent security vendors
Agents act, so the risk moves from what a model says to what it does: the tools it calls, the MCP servers it trusts and the data it can reach. 35 of 56 vendors document agent security. 16 mention MCP servers or gateways by name, and 14 pair agent controls with both runtime guardrails and red teaming.
35 vendors that document agent security
Checked 11 Nov 2026| Vendor | What it documents | Also covers |
|---|---|---|
| A10 Networks TrojAI | TrojAI Defend for MCP discovers servers/tools and enforces communication policies. 1 | Guardrails, Red teaming |
| Adversa AI | Analyzes chains of coding-agent actions and blocks dangerous behavior before completion. 1 | Red teaming |
| Arthur | Agent Behavioral Analytics detects anomalies and indicators of compromise. 1 | Guardrails, AI-SPM |
| Aurascape | Governs MCP activity and tool use with a Zero-Bypass MCP Gateway. 1 | Guardrails, Red teaming, AI-SPM, Data protection |
| Check Point AI Security | Prevents tool/privilege abuse and blocks unsafe autonomous actions. 1 | Guardrails, Red teaming |
| Coralogix AI Guardrails | Supports customized guardrail rules per agent and continuous AI interaction monitoring. 1 | Guardrails, Data protection |
| Credo AI | Agent registry with agent cards and dependency maps; an agentic risk library covers tool misuse, scope drift and inter-agent risk. 1 | AI-SPM |
| CrowdStrike Falcon AI security | Agentic Identity Provider establishes verifiable agent identities and continuous access authorization. 1 | AI-SPM, Model scanning |
| Cyera AI Guardian | Maps each agent’s links to models, tools, MCP servers and data stores; can quarantine agents or block risky actions. 1 | Red teaming, AI-SPM, Data protection |
| DeepKeep | AI Agent Scanner scans agent attack surfaces. 1 | Guardrails, Red teaming, Model scanning |
| Dreadnode | Combines task evaluations, LLM judges, and adversarial testing to assess agent behavior. 1 | Red teaming |
| Enkrypt AI | Offers an MCP scanner and MCP gateway. 1 | Guardrails, Red teaming |
| F5 AI Guardrails | Audits and blocks unauthorized tool calls and agent actions. 1 | Guardrails, Red teaming |
| Fortinet FortiAIGate | Blocks harmful agent tool calls and inspects MCP tool traffic. 1 | Guardrails, Data protection |
| Giskard | Provides agent go/no-go reports and retests fixes through remediation. 1 | Red teaming, Data protection |
| Gray Swan | Supports AI agents with tool access and MCP/tool integrations. 1 | Guardrails, Red teaming |
| Harmonic Security | Governs autonomous workflows at the MCP and tool layers. 1 | AI-SPM |
| HiddenLayer | Protects agentic systems and MCP interactions. 1 | Guardrails, Red teaming, Model scanning |
| Holistic AI | Guardian Agents supervise and intervene in autonomous agent actions. 1 | Red teaming, AI-SPM |
| Knostic | Kirin blocks prompt injection and rogue actions inline before execution. 1 | AI-SPM, Model scanning |
| Microsoft AI security | Task adherence API detects misaligned, unintended, or premature agent tool use. 1 | Guardrails, AI-SPM, Data protection |
| Mindgard | Assesses tool-using agents and systems connected through MCP or A2A servers. 1 | Red teaming, AI-SPM |
| Netskope | Agent Action Control applies intent-, risk-, and access-based policies to agent actions. 1 | Guardrails, Red teaming |
| Noma Security | Approves, reviews, or blocks agents, MCP servers, and skills; scopes allowed actions. 1 | Guardrails, Red teaming, Model scanning |
| NVIDIA NeMo Guardrails | Validates tool inputs and outputs around tool invocation. 1 | Guardrails |
| Oligo Security | Prevents AI agents from making unvetted outbound API calls. 1 | AI-SPM, Model scanning |
| Operant AI | Agent Protector detects threats and blocks zero-click exploits, data exfiltration, and rogue agents. 1 | Guardrails, AI-SPM |
| Palo Alto Networks Prisma AIRS | Includes agent identity and endpoint security modules. 1 | Guardrails, Red teaming, AI-SPM, Model scanning |
| Proofpoint AI Security | Inspects intent and behavior across multi-step agent transactions. 1 | Guardrails, AI-SPM, Data protection |
| SentinelOne Prompt Security | Discovers MCP servers and agents, enforces least privilege, and logs agent actions. 1 | Guardrails, Red teaming, AI-SPM |
| Snyk Agent Scan | Detects malicious skills and security vulnerabilities before installation. 1 | Model scanning |
| Straiker | Runtime response includes an agentic kill switch. 1 | Guardrails, Red teaming, AI-SPM |
| Upwind | MCP Security traces agent prompts, decision chains, tool calls, file changes and downstream API calls end to end. 1 | Red teaming, AI-SPM, Data protection |
| WitnessAI | Enforces approved MCP server and tool access across agents and IDEs. 1 | Guardrails, Red teaming, AI-SPM |
| Zenity | Governs every agent-to-MCP connection at the tool level and evaluates each agent action against policy. 1 | Guardrails, AI-SPM |
What to ask before you shortlist
- Does it see your agents? Discovery comes first. Vendors that inventory agents and MCP servers are on the AI-SPM list.
- Does it enforce, or only observe? Look for inline blocking of tool calls and outbound API calls, not only logs.
- Which coding agents does it cover? Adversa AI, Cato Networks AI Security, Cyera AI Guardian, Knostic, SentinelOne Prompt Security, Snyk Agent Scan, Tenable One AI Exposure and Zenity name Claude Code, Cursor or GitHub Copilot support.
- Who owns the product now? Several agent security startups were acquired in 2025 and 2026. The ledger shows who bought whom.