AI agents outnumber humans, forcing IAM overhaul
The gist
AI agents now vastly outnumber humans in enterprise networks, forcing organizations to rip up old playbooks and scramble for next-gen identity and access controls.
What to know
- By mid-2026, enterprises are juggling over 100 autonomous AI identities per human, exposing traditional IAM weaknesses and driving a rapid shift to cryptographic, context-aware authorization.
- Only 29% of organizations have standardized trust frameworks in place, fueling a boom in security-by-design platforms like Microsoft Agent 365, WSO2 AI Gateway, and Palantir’s Agentic Runtime.
- Regulators like NIST are racing to define AI agent security standards, but just 12% of enterprises fully understand sovereign AI risks amid a complex patchwork of global rules.
IAM’s Human Era Ends
Legacy identity systems collapsed as AI agents exploited static credentials and outdated trust models, exposing enterprises to invisible privilege creep and high-profile breaches.
By late 2025, it became clear that traditional human-centric identity and access management (IAM) systems were fundamentally ill-equipped to handle the surge of autonomous AI agents within enterprises. As Jack Hirsch observed, over 90% of organizations had deployed AI agents, yet only about 10% had any governance strategy in place, exposing a glaring security and identity management gap. Conventional IAM methods relying on static credentials or OAuth grants either offload critical authorization decisions to end users or lack the necessary granularity, underscoring the urgent need for new open standards and architectures tailored specifically to non-human identities.
High-profile incidents such as the 2025 ServiceNow Now Assist vulnerability, which allowed unauthenticated attackers to impersonate any user and bypass MFA and SSO controls, starkly revealed the shortcomings of legacy IAM models that depend on easily compromised trust assumptions like email-based identity and hardcoded secrets. This flaw, carrying a CVSS score of 9.3, catalyzed federal attention with NIST launching a targeted RFI to develop actionable best practices for securing autonomous AI agents, emphasizing the necessity of cryptographic identity verification and permission scoping to prevent similar breaches.
As autonomous AI agents proliferate, their unique behaviors—authenticating, assuming roles, and calling APIs like human users—create invisible privilege creep and untraceable actions if treated as mere application features. Experts like Mrinal Wadhwa argue that agent identity is an infrastructure problem requiring cryptographic proofs for authentication, authorization, and attribution, moving beyond static roles to session-based, risk-aware permissions that grant just-in-time, purpose-bound access. This shift demands continuous, context-aware authorization, immutable logging, and comprehensive visibility across SaaS, cloud, and enterprise environments to effectively manage the rapidly expanding population of non-human identities, which now outnumber humans by over 100 to one.
By mid-2026, the security landscape had dramatically evolved with a surge in AI agent-related incidents, including agent-in-the-middle attacks exploiting trust protocols and malicious code deployment via AI skills, as documented by OWASP and Trustwave’s SpiderLabs. These developments highlight the inadequacy of legacy controls designed for human operators, which collapsed once agents gained the ability to influence their own execution environments. Consequently, the industry faces an urgent imperative to adopt new governance architectures that integrate IAM, privileged access management, identity threat detection, and continuous monitoring to detect improper AI agent behavior, shifting the security focus from merely keeping bad actors out to actively managing autonomous insiders within enterprise ecosystems.
Security-by-Design Becomes Law
With only a fraction of organizations achieving trust maturity, cryptographic agent identities and real-time auditability are now mandatory for AI deployment at scale.
Despite the rapid adoption of autonomous AI agents, enterprises have struggled with low maturity in trust and governance frameworks, with only 29% having standardized models and a trust index barely above 2.4 out of 5. Data provenance and protection have emerged as foundational priorities, as 72% of AI professionals emphasize securing data as the lifeblood of AI, yet complexities in regulatory compliance and SaaS data silos continue to challenge effective governance implementation.
Security-by-design has become the cornerstone of trustworthy AI agent deployment, as industry leaders like Microsoft and experts such as Jeetu Patel advocate embedding security and safety as prerequisites rather than afterthoughts. This approach includes assigning unique cryptographic identities to agents, continuous runtime enforcement, and tightly integrated feedback loops between model behavior and product controls, enabling enterprises to move beyond experimentation toward production-scale AI with measurable ROI and robust compliance.
The unpredictable and autonomous nature of AI agents demands governance frameworks that embrace continuous monitoring, runtime policy enforcement, and reversibility mechanisms to manage risks effectively. Thought leaders like Anneka Gupta highlight the necessity of visibility, governance, and an 'undo button' to mitigate unintended consequences, while emerging platforms such as Drata, TRUSTNOW.EKAM, and Collate are pioneering real-time discovery, cryptographic identity management, and tamper-evident audit trails to ensure operational accountability and compliance at scale.
As AI agent deployment scales, enterprises are shifting from fragmented, post-hoc governance to integrated, centralized frameworks that combine security, observability, and compliance within existing enterprise workflows. Microsoft’s Agent 365 and WSO2’s self-managed AI Workspace exemplify this trend by embedding governance into identity, policy enforcement, and runtime controls, while emphasizing human oversight, cross-functional collaboration, and alignment with evolving regulations like the EU AI Act. This holistic evolution transforms trust from a bottleneck into an accelerator for enterprise AI adoption.
AI Security Arms Race
Specialized platforms emerged to combat agent impersonation and privilege escalation, marking a decisive shift to continuous, protocol-agnostic AI governance.
By late 2025, enterprises confronted escalating challenges in authenticating autonomous AI agents amid threats like token theft and MFA bypass, prompting the emergence of specialized platforms such as Astrix’s AI Agent Control Plane and 1Password’s AI Gateway. These innovations enforce least privilege access through short-lived, scoped credentials and enable secure authentication without exposing raw secrets, thereby reducing compliance risks and operational vulnerabilities in complex enterprise ecosystems.
Early 2026 witnessed a surge in integrated AI security solutions exemplified by Varonis’ acquisition of AllTrue.ai and Palantir’s Agentic Runtime framework, both pioneering real-time visibility and multi-layered guardrails for AI systems. Varonis-AllTrue.ai’s platform uniquely addresses shadow AI and enforces least privilege at scale, while Palantir’s approach treats AI agents as first-class security principals, implementing hardened Kubernetes environments, dynamic runtime policies, and lineage tracking to fill critical gaps absent in consumer AI platforms.
The launch of Operant AI’s Agent Protector and Keycard’s acquisition of Anchor.dev in early 2026 further advanced real-time agentic security by enabling continuous discovery, rogue agent detection, and protocol-agnostic governance. These platforms empower enterprises to prevent privilege escalation and data exfiltration while simplifying deployment and management of autonomous coding agents, effectively eliminating the traditional tradeoff between AI autonomy, capability, and security.
By mid-2026, the proliferation of autonomous AI agents—now outnumbering human users by a factor of 45 in some environments—drove the development of comprehensive control planes like Onyx AI and BeyondTrust’s AI Agent Security. These solutions emphasize real-time enforcement of least privilege, pre-action blocking, and full auditability, reflecting a paradigm shift where AI agents become the dominant endpoint actors requiring the same rigorous security oversight historically reserved for humans, as underscored by BeyondTrust CTO Marc Maiffret’s assertion that 'the actor has changed, but our job has not.'
Complementing these advances, platforms like WSO2 AI Gateway introduced unified AI traffic control and lightweight, deterministic guardrails that enforce least privilege and detect risky content in real time without latency penalties. By integrating regex-based pattern matching, PII masking, and URL validation, WSO2 enables enterprises to maintain consistent governance and observability across diverse AI models and autonomous agents, illustrating the maturation of AI agent security into a nuanced, scalable discipline.
Regulators Target Agent Risks
NIST and industry alliances are racing to define technical and organizational standards for AI agents, as sovereignty and compliance reshape deployment strategies.
By early 2026, regulatory bodies like NIST began distinguishing autonomous AI agents from general chatbots, recognizing the unique security risks posed by agents capable of persistent real-world actions. NIST’s Request for Information sought concrete best practices and actionable recommendations to inform forthcoming technical guidelines, urging organizations deploying AI agents to share their security controls by March 9, 2026, to shape these emerging standards.
The imperative of data sovereignty has surged as enterprises and sovereign entities prioritize deploying AI models within controlled environments—on-premises or private clouds—to maintain governance and avoid geopolitical risks. Companies like Mistral AI emphasize customization through fine-tuning and continued pre-training to align AI models with proprietary data and linguistic nuances, reflecting a broader trend where control over AI models, data, and deployment environments is paramount to meet regulatory and sovereignty demands.
Industry collaborations such as NSS Labs with AWS, Microsoft, and F5 have advanced AI security standards that embed governance deeply into risk and compliance frameworks, moving beyond technical fixes to organizational controls. This governance-driven approach is echoed in enterprise platforms like Armor’s Sovereign AI and TRUSTNOW’s EKAM, which provide real-time control, cryptographic identity management, and audit-ready compliance aligned with regulations like the EU AI Act and HIPAA, addressing the growing regulatory and operational complexity of AI agent deployments.
Despite widespread enterprise adoption of AI agents—86% usage—understanding of sovereign AI risks remains limited, with only 12% of organizations fully grasping these challenges. This gap is exacerbated by fragmented regulatory landscapes, particularly between the U.S. and Europe, and by technical and operational barriers such as infrastructure complexity and vendor lock-in. Experts advocate for selective sovereignty strategies that balance risk and cost, emphasizing the need for integrated governance frameworks that ensure control over AI models, data, and operations throughout the AI lifecycle, supported by continuous compliance monitoring and enforceable contractual rights.
Continuous Oversight, Not Checklists
Automated, real-time governance replaces periodic audits as enterprises struggle to track, control, and prove accountability for millions of unpredictable AI actions.
The maturation of enterprise AI governance is marked by a critical shift from static, periodic oversight to continuous discovery, real-time monitoring, and closed-loop enforcement mechanisms that ensure operational accountability. Leaders like Anneka Gupta emphasize the necessity of visibility, governance, and reversibility to manage the amplified unpredictability of autonomous AI agents, while companies such as Microsoft and Onyx have pioneered platforms embedding identity tracking, runtime policy enforcement, and audit trails that provide comprehensive oversight of AI agent actions. This evolution addresses the growing complexity and scale of AI-driven activities, where traditional human-in-the-loop models become impractical, necessitating automated control planes capable of validating legitimacy and swiftly deactivating anomalous behaviors to meet stringent board, auditor, and regulator expectations.
As autonomous AI agents increasingly operate at speeds and complexities beyond human oversight, enterprises face a pressing need to rethink foundational governance constructs such as identity, access control, and threat modeling. Analysts highlight that agents blur traditional boundaries—are they users, service accounts, or something new?—and require novel approaches to logging, approval workflows, and privilege boundaries to prevent operational risks like unauthorized data deletion or policy circumvention. The rapid proliferation of diverse AI agents, exemplified by the rise of autonomous coding assistants constituting over 50% of deployments, underscores the scalability challenges in applying task-specific guardrails and continuous control, with experts warning that least privilege models and static rules are insufficient in dynamic AI environments where agents can improvise and find workarounds.
The governance landscape is evolving toward integrated, evidence-based accountability frameworks that embed continuous proof and operational transparency into AI lifecycles, moving beyond traditional annual audits or policy documents. Platforms like Drata and Microsoft's Foundry exemplify this by generating tamper-evident audit logs and runtime telemetry that transform operational data into verifiable governance evidence, satisfying escalating demands from boards, auditors, and regulators. This continuous control loop—encompassing discovery, monitoring, enforcement, and evidence generation—is now recognized as essential to mitigate existential risks posed by autonomous AI, with industry voices like Shayne Higdon declaring this the 'accountability moment' where governance failures have tangible consequences for leadership and organizational trust.
Mature AI governance increasingly demands cross-functional organizational coordination and the embedding of security and compliance by design, as highlighted by Dataiku’s Kurt Muehmel and Microsoft’s Secure Future Initiative. This includes clear ownership of AI agent deployment, monitoring, escalation, and accountability across the C-suite—CIOs, CTOs, Chief Data Officers, CFOs, and business leaders—to ensure that AI systems operate within defined boundaries while remaining aligned with business policies. The industry is moving toward 'governed autonomy,' where AI agents orchestrate actions under continuous observability and auditability, supported by modular AI shields and unified control planes that balance operational efficiency with compliance, cost, and carbon management, reflecting a holistic approach to managing the new autonomous AI workforce.
Sovereign AI Shapes Ecosystems
Massive public and private investments are driving centralized, trust-based agent networks and hybrid sovereignty models, making digital sovereignty a core architectural battleground.
By early 2026, AI adoption is increasingly embedding into established trust networks rather than existing as standalone applications, with creators deploying autonomous agents as trusted extensions within messaging platforms, education systems, and small business infrastructures. This evolution into an agent economy leverages pre-existing human relationships to distribute intelligence, underscoring that while AI intelligence may be universal, trust remains inherently local and pluralistic across ecosystems.
The concept of sovereign AI has rapidly expanded beyond simple data residency to encompass comprehensive national infrastructure strategies, as evidenced by massive initiatives like the U.S. Stargate project with its $500 billion commitment, Europe’s €12 billion investment post-2025 Berlin Summit, and Saudi Arabia’s HUMAIN project partnering with AMD on multiple 200 MW data centers. These large-scale sovereign AI efforts highlight a shift toward hybrid sovereignty models that integrate compute, data centers, and AI stacks within geopolitical ecosystems, necessitating multi-dimensional governance frameworks to manage complex AI agent populations across borders.
Enterprises are adopting systemic, centralized governance models to manage AI agents at scale, moving from isolated AI assistants to hub-spoke architectures that centralize model catalogs and API management for cost control, sovereignty, and consistency. This approach addresses challenges such as decentralized model usage, while digital sovereignty becomes a critical architectural concern, especially for European organizations balancing stringent regulatory requirements like the EU AI Act—which 51% of respondents in Omdia’s 2026 report cite as a key driver for sovereign AI investments—with reliance on global cloud platforms.
The hybrid future of enterprise AI sovereignty is characterized by selective implementation strategies that balance control, compliance, and innovation amid fragmented global regulations and vendor dependencies. As Eric Helmer of Rimini Street emphasizes, the transition from AI experimentation to production projects demands robust sovereignty models integrated with existing enterprise systems to navigate cross-border data flows and avoid compliance pitfalls, vendor lock-in, and unpredictable government interventions, thereby sustaining secure, resilient AI operations in complex multinational environments.


















