THE FULL AI SECURITY LANDSCAPE
56 companies, what each one actually does, and whether you would run it alongside Mountain Theory or instead of it. Most of this market is complementary. 41 of these work alongside us, 9 are genuine head-to-heads, and 6 do a different job entirely.
Mountain Theory sits at the execution layer: it checks the action an AI is about to take against your policy and returns ALLOW, HOLD or BLOCK. Almost nothing else in this table does that, which is why so many of these pair with us rather than compete.
Showing 56 of 56 companies
| Company | What they do | With Mountain Theory |
|---|---|---|
| ZenityInline agent runtime security | Controls agent actions inline inside Microsoft Foundry, Copilot Studio, AWS Bedrock and OpenAI AgentKit. | Head to headRarely both. If all your agents live in those platforms Zenity covers them; Mountain Theory reaches the custom and on-prem agents that never touch those control planes. |
| Capsule SecurityAI agent runtime security | Open-source ClawGuard checkpoint that assesses agent intent before a tool call. | Head to headStart with ClawGuard if you are a developer team. Move to Mountain Theory when policy needs an owner outside engineering and an auditor needs evidence. |
| Eve SecurityAgentic AI intent and policy | Infers intent across multi-step reasoning chains to detect goal drift, with Agent-in-the-Loop review. | Head to headBoth check before the action. Eve reasons probabilistically about intent; Mountain Theory checks the action deterministically so the answer is auditable. |
| Sondera.aiDeterministic policy enforcement | Open-source harness that compiles natural-language rules into formally verified Cedar policy and enforces outside the model. | Head to headClosest philosophy in the market. Sondera for coding agents inside dev frameworks; Mountain Theory for production agents across the business. |
| Geordie AIAI agent security and governance | Maps the agentic footprint across an enterprise and applies automated mitigations at platform level. | Works alongsideGeordie tells you where your agents are. Mountain Theory governs what they do. Discovery feeds enforcement. |
| StraikerAgentic AI security | Runtime protection for AI agents, aimed at enterprises and frontier labs. | Head to headSimilar problem statement. Evaluate on whether control is model and framework agnostic and whether policy is written by the risk owner. |
| Trent AIAI agent security | Security for autonomous AI agents. London-based, backed by OpenAI, Spotify and Databricks operators. | Head to headSame category, earlier stage. Worth watching rather than displacing today. |
| MultifactorMulti-agent security controls | Verifiable, fine-grained controls for complex multi-agent systems, built for CISOs. | Head to headOverlapping intent on multi-agent estates. Very early. |
| Operant AIMCP and runtime security | Runtime security for AI applications and MCP deployments. | Works alongsideOperant secures the protocol path. Mountain Theory governs the action once a tool call is made, whatever route it arrived by. |
| RunlayerMCP security | Security for MCP server deployments. First mover on the protocol. | Works alongsideRun both if MCP is central. They harden the connection, we govern the action it enables. |
| Astrix Security (Cisco)Non-human identity | Discovers and governs API keys, service accounts, OAuth tokens and AI agent identities. Acquired by Cisco. | Works alongsideRun both. Astrix establishes who the agent is and that its credentials are sound. Mountain Theory decides whether the action it is attempting is allowed right now. |
| SaviyntAI identity and access management | Extends enterprise IAM to AI agents with an identity control plane. | Works alongsideRun both. Saviynt authorises the identity at session start; Mountain Theory authorises every action after it. |
| Oasis SecurityNon-human identity management | Machine identity lifecycle: discovery, governance and remediation of NHIs. | Works alongsideRun both. Credential hygiene lowers how often a wrong actor acts; execution control decides whether the act goes through. |
| WideField Security (Cisco)Identity lifecycle for the agentic SOC | Identity, credential and session intelligence feeding Splunk investigations. Acquired by Cisco. | Works alongsideSequential. WideField helps the SOC understand an agentic incident; Mountain Theory reduces how many reach the SOC. |
| HiddenLayerAI model and agentic runtime security | Scans model artifacts across 35+ formats and defends at inference without needing weights, training data or prompts. | Works alongsideRun both. HiddenLayer tells you the model is trustworthy. Mountain Theory governs what a trustworthy model then does. |
| Noma SecurityUnified AI and agent security | AI asset discovery, data context, posture management and an agentic risk map. | Works alongsideRun both in a large estate. Noma shows exposure; Mountain Theory intercepts the transaction. |
| Protect AI (Palo Alto)ML security platform | Model scanning, posture, red teaming and runtime, now the core of Prisma AIRS. | Works alongsideNow part of Palo Alto. See the Prisma AIRS entry. |
| Robust Intelligence (Cisco)AI firewall and red teaming | AI Firewall for real-time model protection, now Cisco AI Defense. | Works alongsideContent and model protection. Mountain Theory covers the action layer neither addresses. |
| Oligo SecurityApplication-level AI integrity | eBPF kernel-level runtime monitoring of AI workloads, catching library-level exploits. | Works alongsideDifferent depth of the stack. Oligo watches the workload; Mountain Theory gates the action. |
| MindgardAI red teaming | First DAST for AI. Continuous automated red teaming with a large attack library. | Works alongsideTesting, not enforcement. Mindgard finds the weakness; Mountain Theory is what stops it being exploited in production. |
| TrojAI (A10 Networks)AI red teaming and runtime protection | Red teaming that probes models, agents and applications at build time, plus real-time threat protection at runtime. Acquired by A10 Networks, announced 15 June 2026, to support sovereign AI security. | Works alongsideBuild-time assurance plus their own runtime layer, now inside a network infrastructure vendor. Their runtime defends the model and the application. Mountain Theory governs the action a clean model triggers, so the two sit at different points and most estates would run both. |
| Haize LabsAI safety ratings | Red teaming that produces safety ratings, working with frontier labs. | Different jobRatings and benchmarks. Different buyer, different purpose. |
| NoveeAutonomous AI pen testing | Autonomous black-box red teaming with a proprietary AI attacker. | Different jobOffensive testing. Complements any defensive control including ours. |
| SPLX (Zscaler)AI runtime protection and guardrails | Inline prompt and output filtering, natural-language policy, red teaming, now inside Zscaler. | Works alongsideRun both. SPLX stops the model saying something unsafe; Mountain Theory stops the agent doing something unsafe. Different failure. |
| Prompt Security (SentinelOne)GenAI security | Inspects every prompt and response for DLP, injection and jailbreaks. Now in Singularity. | Works alongsideContent safety. Sits before the action layer, not on it. |
| Lakera (Check Point)GenAI security | Real-time protection from prompt injection, data leakage and toxic content. | Works alongsidePrompt-layer defence. Mountain Theory is the backstop for when the prompt filter is beaten. |
| Guardrails AIOpen-source LLM guardrails | Programmable guardrails on LLM outputs, developer-first. | Works alongsideOutput validation. Pairs with, does not substitute for, execution control. |
| Virtue AIAI security and compliance | Multi-modal guardrail suite with continuous model benchmarking, strong academic pedigree. | Works alongsideContent and model assurance alongside action control. |
| Galileo AIAI evaluation platform | Evaluation foundation models detecting hallucination, injection, PII and toxicity in real time. | Works alongsideOutput quality and safety. Different question from whether an action should run. |
| Patronus AILLM evaluation and testing | Automated evaluation detecting LLM mistakes at scale. | Different jobEvaluation tooling. Rarely in the same procurement. |
| Arthur AIAI monitoring and governance | Monitoring, evaluation and governance with an LLM firewall, pivoting to agentic governance. | Works alongsideObservability plus governance. Mountain Theory supplies the enforcement they describe. |
| AurascapeAI-native security layer | Real-time visibility and intent-based controls across thousands of AI applications. | Works alongsideBreadth across AI app usage. Complements depth at the execution boundary. |
| Promptfoo (OpenAI)LLM security testing | Open-source CLI for adversarial testing of LLM apps. Acquired by OpenAI. | Different jobDeveloper testing tool. Not an enforcement product. |
| Lasso SecurityLLM security and MCP gateways | Shadow AI discovery, data-flow monitoring and MCP gateways, distributed via Cloudflare. | Works alongsideRun both. Lasso controls the user-to-model boundary; Mountain Theory governs what an agent already inside your systems does. |
| ThalesAI runtime data security | AI Security Fabric controlling which data agents may access, with an MCP gateway on the 2026 roadmap. | Works alongsideRun both in regulated estates. Thales is the data vault; Mountain Theory is the decision gate. |
| CalypsoAI (F5)Inference security | Red teaming and real-time threat defence at the inference layer. Acquired by F5. | Works alongsideInference-layer defence, complementary to action control. |
| Nightfall AICloud-native DLP for AI | ML data discovery, classification and protection across SaaS, APIs and browsers. | Works alongsideData loss prevention. Different object entirely. |
| LiminalSecure multi-model AI access | Intelligent data masking rather than redaction, single-tenant, for regulated industries. | Works alongsideProtects the data in the prompt. Mountain Theory protects the system from the action. |
| WitnessAIAI governance and security | Visibility, control and compliance for enterprise AI usage. | Works alongsideUsage governance alongside action enforcement. |
| Harmonic SecurityAI data loss prevention | DLP purpose-built for AI workflows rather than retrofitted. | Works alongsideData-layer control, complementary. |
| Pillar SecurityAI lifecycle security | Discovery, testing and adaptive guardrails across the AI lifecycle. | Head to headBreadth versus depth. Pillar covers the lifecycle; Mountain Theory is deeper at the one boundary that stops an action. |
| Credo AIAI governance platform | AI governance, risk and compliance programme management. | Works alongsideRun both. Credo produces the policy and the paperwork; Mountain Theory is what actually enforces it. |
| Palo Alto Networks (Prisma AIRS)Platform AI security | Model scanning, posture, red teaming and runtime across the AI lifecycle, from the Protect AI acquisition. | Head to headOften the default if you already run Palo Alto. Ask what stops a custom agent on your own infrastructure tonight. |
| Cisco (AI Defense)Platform AI and identity security | AI Defense plus Astrix and WideField, building an identity-led trust layer for agentic AI. | Works alongsideStrong on identity and investigation. Mountain Theory supplies inline action enforcement. |
| MicrosoftPlatform AI security | Security Copilot, Agent 365 control plane and an agent governance toolkit. | Works alongsidePlatform-native controls for Microsoft-hosted agents. Mountain Theory reaches everything else. |
| GoogleCloud AI security | AI-SPM and AI protection through Wiz, plus the A2A protocol. | Works alongsideCloud-native posture. Complementary to execution control. |
| SentinelOneEndpoint and AI security | Singularity platform with Prompt AI Agent Security from the Prompt Security acquisition. | Works alongsideEndpoint and content security. Different layer. |
| CrowdStrikeEndpoint and agent security | Falcon and AIDR, expanding into AI identity through SGNL and Pangea. | Works alongsideEndpoint and identity strength. Mountain Theory governs the action. |
| Check PointNetwork and AI security | Infinity AI with Lakera for AI-native detection. | Works alongsideNetwork and prompt layers, complementary. |
| DarktraceAI anomaly detection | Self-learning anomaly detection moving into AI tool security. | Works alongsideDetection and anomaly. Mountain Theory is prevention at the action. |
| UpwindRuntime-first CNAPP | Runtime cloud security with AI posture, inventory and behaviour tracing via eBPF. | Works alongsideRun both. Upwind explains what happened across your cloud; Mountain Theory decides whether it happens. |
| Apex Security (Tenable)AI visibility and policy | AI activity visibility and policy enforcement. Acquired by Tenable. | Works alongsideVisibility layer, complementary. |
| PindropVoice biometric AI security | Deepfake detection and authentication for the voice channel. | Works alongsideRun both if you operate voice AI. They authenticate one channel; we govern action across all of them. |
| ZeroDriftAI communications compliance | Checks every AI-generated message against SEC, FINRA, HIPAA and firm policy before it sends. | Works alongsideRun both in financial services. ZeroDrift governs what AI says; Mountain Theory governs what it does. |
| 7AIAgentic SOC automation | Autonomous AI agents automating security operations work. | Different jobAI for security rather than security for AI. Notably, their agents are themselves something you would want governed. |
| ArmadinAutonomous AI SecOps | Agentic security operations, founded by the Mandiant founder. | Different jobSame note: an autonomous SecOps agent is a strong candidate for execution-layer control. |
No company matches that search.