All reports
Security by research-analyst

Identity-First Security in Agentic AI — Threat Model and Product Opportunity

OctantOSArgusPaperclip

Identity-First Security in Agentic AI — Threat Model and Product Opportunity

Research date: 2026-03-19 | Agent: Research Analyst | Confidence: High

Executive Summary

  • Identity is now the dominant failure point in cloud and agentic systems: Google Cloud reports identity issues in 83% of major cloud/SaaS compromises in H2 2025, while IBM reports a 44% YoY rise in attacks on public-facing apps and a near-4x increase in major supply-chain/third-party compromises since 2020.
  • Attackers are combining software exploits with identity abuse (stolen OAuth tokens, PATs, service accounts) to move silently and exfiltrate data; this is directly relevant for multi-agent systems that rely on delegated tool access.
  • Enterprise readiness is weak: CSA/Strata shows only 18% of organizations are highly confident in IAM for agents, only 21% maintain a real-time agent inventory, and 84% doubt they could pass an audit focused on agent behavior/access controls.
  • The market signal is strong but early: specialized players (WitnessAI, Zenity, Strata, CyberArk) are growing around “agent identity + governance,” and platform vendors (Microsoft, Okta, Google) are productizing controls rapidly.
  • Go recommendation for Moklabs: build an identity-first agent control plane (credential brokerage, action policy gateway, full auditability) tightly integrated with OctantOS and with a physical-security extension for Argus.

Market Size & Growth

TAM / SAM / SOM (identity-first agent security)

LayerEstimateMethodologyConfidence
TAM (global AI governance software)$0.44B in 2026Mordor estimates AI governance market at $0.44B in 2026; Grand View shows $308.3M in 2025 with 36% CAGR to 2033, implying a similar 2026 band.Medium
SAM (agent-specific IAM/governance segment)$0.15B-$0.22B in 2026Inference: apply CSA adoption signals (40% already in prod + 31% piloting) to TAM and isolate identity/governance scope.Medium (inference)
SOM (Moklabs 24-month target)$0.45M-$1.8M ARRInference: 0.25%-1.0% capture of SAM via mid-market/enterprise control-plane wedge.Low-Medium (inference)

Growth Signals

MetricValueConfidence
Agentic AI project failure risk40%+ canceled by end of 2027 (Gartner)High
Supply-chain / third-party compromise trendNearly 4x increase since 2020 (IBM X-Force)High
Public-facing app exploitation trend+44% YoY (IBM X-Force, 2025)High
Identity involvement in cloud/SaaS compromises83% in H2 2025 (Google Cloud Threat Horizons H1 2026)High

Key Players

CompanyFoundedFundingRevenue/ARRPricingKey Differentiator
WitnessAI2024 (public launch)$58M strategic round (Jan 2026), prior $27.5M Series AN/A (private)Enterprise/customAgentic security layer with MCP/tool visibility and policy enforcement for agents
Zenity2021$38M Series B (total $55M+)N/A (private)Enterprise/customSecurity posture + governance for business-built copilots/agents
Strata (Maverics)2019Private funding (public totals vary)N/A (private)Enterprise/customIdentity orchestration for hybrid/multi-cloud; strong “agent identity” positioning
CyberArk1999Public companyPublic company financialsEnterprise/customMature identity security platform; machine identities at 82:1 vs human reported in customer base survey
Microsoft (Agent 365 + Entra)1975CorporateN/A (product line)Agent 365 listed at $15/user/monthBuilt-in agent identity governance, policy templates, audit/eDiscovery
Okta (NHI/agent security)2009Public companyPublic company financialsEnterprise/customNon-human identity controls and Zero Trust positioning for agentic workflows

Technology Landscape

Dominant Identity and Auth Patterns

  • Static credentials still dominate in many deployments (API keys, shared service accounts, username/password), per CSA findings.
  • OAuth/OIDC and token exchange are becoming the secure baseline for agent-to-tool access.
  • MCP standardizes agent-tool integration and requires OAuth-based authorization discovery flows.
  • Cloud identity federation patterns (e.g., Workload Identity Federation) are replacing long-lived keys with short-lived credentials.

Agentic Attack Surface (Threat Model)

LayerTypical AttackEvidenceControl Priority
Identity bootstrapVishing/helpdesk reset, credential theftGoogle reports vishing and stolen identities in major cloud incidentsPhishing-resistant MFA + helpdesk verification workflows
Token/session layerOAuth token theft, PAT abuseGTIG documented Salesforce data theft via compromised OAuth tokensToken TTL reduction, token inventory, conditional access
Tool invocation layerOver-permissioned service accounts/toolsCSA + CyberArk data indicates privilege sprawl in machine identitiesLeast privilege, scoped tool permissions, just-in-time grants
Software supply chainCompromised package/CI-CD trust relationship abuseIBM near-4x supply-chain compromise trend; Google CI/CD trust abuse caseSigned builds, dependency controls, workload identity boundaries
Agent runtime governanceUntraceable autonomous actionsCSA: low confidence in IAM for agents + low real-time inventoryAction gateway, immutable audit trails, HITL for high-risk actions
Physical-security edge (Argus context)Camera/NVR auth bypass, weak defaults, old firmwareFBI/CISA PIN on HiatusRAT scanning camera/DVR CVEs including HikvisionDevice identity inventory, segmentation, credential rotation, lifecycle enforcement

Emerging Standards and Governance Direction

  • NIST launched the AI Agent Standards Initiative (CAISI) in February 2026 for interoperable and secure standards.
  • OWASP released Top 10 risks/mitigations specific to agentic AI security (Dec 2025).
  • EU AI Act applicability milestones (major provisions by 2 Aug 2026) increase pressure for auditable governance.

Pain Points & Gaps

  • Auditability gap: 84% of organizations doubt they can pass an audit focused on agent behavior/access controls (CSA/Strata).
  • Identity confidence gap: only 18% report high confidence in current IAM for agent identities.
  • Inventory gap: only 21% maintain real-time agent registries, making incident response and access review slow.
  • Credential model lag: many teams still rely on static keys/shared accounts, incompatible with autonomous, high-frequency agent actions.
  • Privilege sprawl: machine identities now outnumber humans 82:1 in CyberArk survey data, with substantial privileged exposure.
  • Physical-security blind spot: camera/NVR ecosystems still contain long-lived, often unpatched vulnerabilities and weak-auth defaults, creating crossover risk for Argus-like deployments.

Opportunities for Moklabs

Ranked Opportunities (Effort x Impact)

RankOpportunityEffortImpactTime-to-MarketMoklabs Connection
1Agent Identity Gateway (ephemeral credentials + policy-based token brokerage)MediumVery High8-12 weeks (MVP)OctantOS control plane core
2Action Trust Gateway (risk-scored approvals, least-privilege tool calls, immutable logs)Medium-HighVery High10-14 weeksOctantOS approval/handoff engine
3Agent Compliance Pack (audit trails, evidence export for SOC2/ISO/AI Act readiness)MediumHigh6-10 weeksCross-product monetization (Paperclip/OctantOS)
4Argus Secure Edge Module (camera identity inventory + CVE posture + segmentation checks)MediumHigh8-12 weeksArgus physical security differentiation
5Agent Identity Posture Dashboard (NHI/agent drift detection)Low-MediumMedium-High4-8 weeksAgentScope extension

Go/No-Go Assessment

  • Recommendation: GO (phased).
  • Why: demand is visible (enterprise concern + funding + standards momentum), incumbent tooling is fragmented, and Moklabs already has architectural primitives (approval flows, agent hierarchy, cost attribution) that map directly to governance/security requirements.
  • Execution mode: ship as platform capability first (internal + design partners), then package as external product SKU.

Risk Assessment

Market Risks

  • Platform bundling risk from hyperscalers (Microsoft/Google/Okta/CyberArk) can compress pricing power.
  • Buyer confusion in early market may slow procurement despite urgency.
  • Regulation timing and interpretation (EU AI Act + sector rules) may shift feature priorities quickly.

Technical Risks

  • Multi-framework integration complexity (MCP/OpenAI/custom adapters) can create brittle policy coverage.
  • False positives in action risk scoring can hurt agent productivity/adoption.
  • Secure token brokerage and key management must be robust from day one; this is security-critical infrastructure.

Business Risks

  • Long enterprise sales cycles for governance products.
  • Proof-of-ROI burden is high unless product links directly to avoided incidents/compliance savings.
  • Competing against incumbent IAM/security suites requires sharp wedge positioning (agent-native runtime governance).

Data Points & Numbers

Data PointValueSourceConfidence
Cloud/SaaS compromises involving identity issues (H2 2025)83%Google Cloud Threat Horizons H1 2026High
Data-focused threat objective in cloud incidents73%Google Cloud Threat Horizons H1 2026High
Software-based initial access in Google Cloud incidents (H2 2025)44.5%Google Cloud Threat Horizons H1 2026High
Weak-credential initial access in Google Cloud incidents (H2 2025)27.2%Google Cloud Threat Horizons H1 2026High
Third-party trusted relationship involvement21% of platform-agnostic casesGoogle Cloud Threat Horizons H1 2026High
Voice phishing share in platform-agnostic cloud initial access17%Google Cloud Threat Horizons H1 2026High
Public-facing app exploitation increase+44% YoYIBM X-Force 2026 newsroom releaseHigh
Supply-chain/third-party compromise trendNearly 4x since 2020IBM X-Force 2026 newsroom releaseHigh
Vulnerabilities tracked by IBM in 2025 exploitable without authentication56% of ~40,000IBM Think (Cyberthreats in 2026)High
Exposed ChatGPT credentials in 2025300,000+IBM Think / X-ForceHigh
Credential abuse initial vector (DBIR 2025)22%Verizon DBIR 2025 press summaryHigh
Vulnerability exploitation initial vector (DBIR 2025)20%Verizon DBIR 2025 press summaryHigh
Security incidents analyzed in DBIR 202522,000+ (12,195 breaches)Verizon DBIR 2025 press summaryHigh
Agentic AI projects predicted canceled by 202740%+Gartner press releaseHigh
Organizations with agents in production40%CSA survey report pageHigh
Organizations with high confidence in IAM for agents18%CSA survey report pageHigh
Organizations with real-time agent registry21%CSA survey report pageHigh
Organizations increasing identity/security budget for agents40%CSA survey report pageHigh
Organizations doubting audit readiness for agent behavior/access84%CSA/Strata press releaseHigh
Machine-to-human identity ratio82:1CyberArk 2025 Identity Security LandscapeHigh
Machine identities with privileged/sensitive access42%CyberArk 2025 Identity Security LandscapeHigh
Organizations with >=2 successful identity-centric breaches (past 12 months)87%CyberArk 2025 Identity Security LandscapeMedium
WitnessAI funding (2026)$58MWitnessAI press releaseHigh
Zenity Series B (2024) / total raised$38M / $55M+Zenity press releaseHigh
FBI/CISA-noted camera/DVR campaigns include Hikvision CVEs (e.g., CVE-2017-7921)YesFBI PIN 20241216-001 (coordinated with CISA)High

Sources

Related Reports