Map / Model Security
Model Security
Secures the model itself across the ML lifecycle — supply-chain scanning of model files for backdoors, malware, and poisoning. Runtime protection against adversarial inputs and integrity drift in production.
Quick answer: The map tracks 57 companies in Model Security; 27 list it as their primary category. The most heavily funded are HiddenLayer ($155M), ActiveFence ($140M) and Alice ($140M). Most are based in USA (28), Israel (10) and UK (4). 8 have been acquired.
57 companies, A–Z
- ActiveFenceAI safety, security and trust: red teaming, content/threat detection (disinformation, hate, fraud, CSAM), GenAI guardrails, model risk evaluation.
- AdvaiAI testing and monitoring platform for secure, compliant AI deployment; adversarial robustness testing; continuous monitoring of deployed models; AI governance and assurance workflows; targets UK government and regulate…
- Adversa AIAI security platform for red teaming AI agents, GenAI applications, and ML models; comprehensive adversarial testing across the AI lifecycle; threat modeling for AI systems; security research and publications informing…
- AI AramidEnterprise AI security control plane sitting between applications and every LLM (cloud or self-hosted), controlling every request before it leaves the network: a traffic gateway, real-time threat blocking, automated red…
- AliceWonderSuite: WonderBuild (pre-launch adversarial stress-testing of models/apps/agents), WonderFence (runtime guardrails), WonderCheck (ongoing automated red-teaming, drift detection).
- AptaSentryGoal-based automated red teaming across text/audio/video/image (AptaRed), agentic workflow stress-testing for prompt injection/privilege escalation/data exfil (AgentRed), model risk evaluation with findings fed back int…
- Arize AIAI/LLM observability and evaluation; tracing, monitoring, and root-cause analysis for LLM and agent apps; LLM guardrails on input (jailbreak) and output (relevance); open-source Phoenix.
- AttestableZero-knowledge proof platform that cryptographically verifies AI system integrity, letting organizations confirm a model's outputs were correctly computed from a specified model and inputs without trusting hardware vend…
- Bosch AIShieldAI security from Bosch across two layers.
- Citadel AIAutomated robustness testing of AI models during training (Citadel Lens); continuous production monitoring for anomalies and data drift without labels (Citadel Radar); compliance reporting for AI standards.
- Cranium AIAgentSensor (agent discovery + tool + agent-to-agent network mapping), CloudSensor (cloud security monitoring), CodeSensor (source-code scanning for models/datasets/AI packages), Arena (agent-based adversarial simulatio…
- DataKryptoFHEnom for AI is a zero-knowledge framework combining DataKrypto's patented high-speed fully homomorphic encryption (FHE) with Trusted Execution Environments to keep data, model weights and queries encrypted across the…
- DeepchecksProduction-grade platform unifying evaluation, observability, testing, and monitoring for AI systems and LLM agents; continuous measurement, tuning, and improvement of AI agents over time; agent behavior validation pre-…
- DreadnodeOffensive AI security platform with Strikes (AI agent training ground for offensive cybersecurity), Spyglass (AI red teaming toolkit for probing deployed AI), and Crucible (AI hacking sandbox); adversarial testing of de…
- Duality TechnologiesSecurePlus privacy-preserving data collaboration platform combining homomorphic encryption with data science; secure multiparty data collaboration using TEEs for AI/ML workloads; AI assistant for sensitive data analysis…
- Dynamo AIEnterprise platform for secure and compliant GenAI infrastructure; DynamoEval for risk assessment; DynamoEnhance for risk remediation; DynamoGuard real-time runtime guardrails; federated learning and differential privac…
- Enkrypt AIEnd-to-end LLM security platform spanning the full GenAI lifecycle: red teaming during development, risk discovery in staging, runtime guardrails in production; automated red teaming across LLMs, Vision-Language Models,…
- EnveilZeroReveal Secure AI for encrypted ML training, inference, and validation across data silos; homomorphic encryption and secure multiparty computation protecting Data in Use; privacy-preserving search and analytics; cros…
- EqtyLabAI Integrity Suite (AI Guardian, Governance Studio) that cryptographically certifies and governs data, models, and agents at runtime.
- Fiddler AILLM observability (hallucination/factuality, toxicity, PII detection, prompt safety, jailbreak attempts, drift), traditional ML model monitoring (drift, accuracy, bias/fairness, explainability), Fiddler Trust Models, Fi…
- GiskardOpen-source Python library for LLM testing + Giskard Hub enterprise SaaS for continuous red teaming, automated vulnerability scanning, hallucination/contradiction detection, RAG evaluation, golden-dataset management, Gi…
- Gray Swan AIAI Security Suite (Cygnal/Cygnet I/O filtering + Shade automated testing), AI red-teaming for frontier labs, Gray Swan Arena (3M+ attack corpus), Circuit Breakers, benchmarks (AgentHarm, HarmBench, WMDP, CyBench, MMLU).
- Haize LabsRuns algorithmic adversarial stress-testing and safety ratings across text, audio, code, video, image, and web modalities.
- HardshellAI security platform protecting training data from poisoning and leakage; adversarial ML defense; data poisoning detection during model training; pre-training data security validation; protection against backdoor attack…
- HiddenLayerAISec Platform 2.0 — AI Discovery (asset inventory, AIBOM, Model Genealogy lineage), AI Supply Chain Security (static scanning of model files for graph-level backdoors, ShadowLogic, control-vector injection, embedded ma…
- HirundoThe first machine unlearning platform, removing unwanted data, behaviors, and vulnerabilities from trained AI models without full retraining.
- I-THAMCybersecurity platform purpose-built for AI, defending models, agents, data, and AI-powered applications with continuous monitoring across the AI stack — from prompt to tensor.
- IrregularFrontier-model security lab; adversarial testing of foundation models for cyber-misuse potential and resilience under attack; works with leading AI labs (OpenAI, Anthropic) and governments.
- JozuJozu Agent Guard — zero-trust AI runtime that executes agents, models, and MCP servers inside secured environments with policy enforcement and guardrails that agents cannot disable; vetting, signing, and governing AI ar…
- LakeraAI-native security platform for agentic AI applications; flagship products Lakera Red (continuous AI red teaming) and Lakera Guard (real-time runtime enforcement); detection rates above 98% with sub-50ms latency and fal…
- LatticaAIPrivacy-preserving AI inference platform using Fully Homomorphic Encryption; encrypted model inputs and outputs across browser, mobile, and server contexts; integration with major LLM providers; FHE-optimized for AI/ML…
- LoricaConfidential AI infrastructure layer enabling privacy-preserving model deployment for AIaaS providers and enterprises; hardware-backed security and attestation at every layer; one-click deployment of confidential model…
- Mirror SecuritySecurity of AI platform with three core technologies: AgentIQ (full-spectrum agentic security with guardrails and compliance), Discover (risk assessment and automated red teaming), and VectaX (Fully Homomorphic Encrypti…
- Mithril SecurityConfidential AI inference allowing models to run on private data inside hardware-based trusted execution environments.
- NeuralTrustComprehensive AI security platform spanning gateway, runtime, and testing: TrustGate (AI gateway with authorization, rate limiting, content filtering across LLM providers); TrustLens (observability and runtime monitorin…
- Noma SecurityEnd-to-end AI lifecycle coverage (discovery → posture → red teaming → runtime), Agentic Risk Map (blast-radius visualization), AI supply-chain security, MLOps tool security, AI threat detection & response, immediate con…
- OpaqueConfidential AI platform enabling enterprises to run cloud-scale general-purpose AI workloads on encrypted data with hardware-attested execution; Confidential Agents for RAG and agentic workflows; Opaque Studio for acce…
- Prediction GuardPrivate, compliant (SOC 2 Type II, HIPAA) LLM hosting and inference with integrated output checks and guardrails via API.
- Prime SecurityAgentic Security Architect — autonomous AI agents conducting security design reviews before code is written; embedded directly in engineering workflows (Jira, etc.); design-flaw detection; near-full coverage of planned…
- Protect AIAI/ML security platform covering the full lifecycle from development to runtime; ML supply chain security (model scanning via ModelScan, dataset and notebook scanning via NB Defense); AI-SPM via Layer; AI red teaming vi…
- Protectt.aiOriginally India's mobile app security and fraud-control platform (RASP, code obfuscation, anti-tampering, device binding, and SIM-swap/transaction-fraud defense for BFSI), Protectt.ai has extended into security for AI…
- Protopia AIStained Glass Transform technology applying mathematical noise to data before AI inference, preserving model accuracy while protecting sensitive content from the model provider; round-trip protection for LLM endpoints;…
- Realm LabsDeep Neural Inspection (DNI) — measures AI's internal mechanisms during inference to observe, detect, and prevent misbehavior before it surfaces as bad output; unified AI Detection & Response platform; signals of halluc…
- Robust IntelligenceEnd-to-end AI security platform protecting AI models and applications across the full lifecycle from development to production; pioneered the industry's first "AI Firewall" — runtime protection against prompt injection,…
- Safe IntelligenceML model validation, robustification, and monitoring platform; formal verification of AI model properties; robustness testing against adversarial inputs; runtime monitoring of deployed models; AI governance and complian…
- SarusPrivacy-preserving SQL/pandas/scikit-learn rewriting engine using differential privacy; computation-to-data architecture leaving data at source; SarusLLM enabling privacy-preserving LLM fine-tuning on sensitive data; de…
- SecretariumKlave for AI confidential computing platform for private, verifiable AI inference on sensitive data; trusted execution environment-based deployment; deterministic and verifiable computation; RAG security; integration wi…
- SecurityBreakAI threat intelligence platform focused on adversarial activity across the AI ecosystem: AI agent monitoring, prompt threat intelligence, and AI supply-chain scanning.
- SkyldAI model protection platform securing on-device ML models from reverse engineering, theft, and tampering; cryptographic obfuscation of model weights and architecture; protection of intellectual property in deployed mode…
- StarseerAI-Verify (pre-deployment model validation via interpretability — detects backdoors, hidden capabilities, misaligned representations); AI-EDR (versioned behavioral detections running continuously across AI endpoints); A…
- StyrkLLM/GenAI guardrails; AI model vulnerability assessment, adversarial robustness testing, bias detection, and privacy protection for AI models.
- TinfoilConfidential AI platform that runs models inside secure hardware enclaves, offering private inference, a private chat and an OpenAI-compatible inference API with zero-trust, zero-access, zero-retention guarantees; colla…
- TrojAITrojAI Detect (pre-deployment red teaming), TrojAI Defend (runtime firewall), Defend for MCP, prompt injection / jailbreak / data leakage / tool exploitation detection.
- TrojanVectorsAutomated red-teaming agents that probe AI systems through simulated adversarial attacks.
- Valarian TechnologiesACRA AI private AI model hosting platform for on-premise deployment in secure environments; zero-trust architecture; tactical/defense-focused AI workloads; air-gapped and disconnected deployment support; broader Valaria…
- Virtue AIAI-native security platform for agentic frameworks and LLM applications; AgentSuite product line covering red teaming, runtime guardrails, and security testing for agent workflows; deep technical research backing includ…
- WoolyAIModel capability monitoring for self-hosted open-weight LLMs.
Other categories
Observability & Governance · Runtime & Guardrails · AI Identity · MCP & LLM Gateways · AI Red Teaming · AI CSPM · AI Data Governance · AI Network Security · AI Endpoint Security · AI Code Security · AI Browser Security · AI SSPM · Sandboxing