Agents as Identities: The IAM Gap Every Enterprise Is Ignoring

Research notes

Todd Thiemann (principal analyst, Omdia) provides one of the clearest frameworks for understanding identity security for AI agents in his TechTarget opinion piece
For years, security controlled risk by controlling identities: employees authenticate through IdPs, service accounts connect systems, API keys let workloads talk to cloud services
AI agents entered enterprises quietly as productivity tools, then got connected to critical business services: Salesforce, Snowflake, GitHub, Jira, production databases, cloud environments
Agents now retrieve information, trigger workflows, update records, write/deploy code across multiple systems — sometimes autonomously, sometimes unclear which
This makes them identities, not just tools — and most enterprises have no security/governance models for them
Thiemann's framework: treat agents as first-class identities with owners, access, behavior, risk, and lifecycle controls

Pattern is consistent across organizations: new identity layer built on existing infrastructure with almost none of the controls IAM teams spent a decade putting in place
An agent might be created by one team, used by another, connected to five different applications, running on credentials provisioned for a completely different purpose
Got broad access early because someone needed it to work and didn't want to slow things down
Result: sprawl of high-privilege, low-visibility actors most security teams can't inventory, let alone govern

Enterprise AI agent fleets have roughly doubled since December 2025 (Gravitee State of AI Agent Security Report, April 2026, n=750 CIOs/CTOs/VPs)
48% of production AI agents are running unsecured (Gravitee)
Only 9.5% of organizations secure more than 80% of their agents (Gravitee)
Mean monitoring coverage: 52%, barely moved since Dec 2025 (46.96% → ~52%) despite fleet doubling
81.7% of organizations plan to deploy more agents in next 12 months, with 28% planning significantly more
Gartner predicts Fortune 500 will have over 150,000 agents by 2028, up from fewer than 15 in 2025
82% of organizations discovered at least one AI agent created without knowledge of security, IT, or governance teams in past year; 41% found this happening multiple times (CSA 2026 survey)

54% of organizations have experienced or suspected an AI agent security/data privacy incident in past 12 months (Gravitee, April 2026)
34.9% confirmed an incident occurred; Telecoms (67.3%) and Financial Services (54.7%) lead sector incidence rates
65% of organizations experienced a security incident involving an AI agent in the past year, with 61% reporting exposure/mishandling of sensitive data (CSA 2026 survey)

Excessive permissions and over-privileged access: most consistently reported failure pattern — agents granted broader access than their function requires, often rooted in shared service accounts or inherited credentials
Data retention and privacy violations: agents storing, caching, or processing data beyond what's needed
Other patterns from Gravitee report include: misconfigured integrations, lack of pre-deployment governance, and deliberate adversarial exploitation (new pattern in April 2026)
Only 19.7% of organizations say all their agents are fully secured and governed before going live
No single pre-deployment control is used by even 40% of organizations: named person accountable (37.8%), security review from IT/CISO (35%), documented process to pause/revoke access (34.1%)

Discovery/Inventory: Need to answer who owns this agent, who can invoke it, what systems connected, what credentials used, what can it read/write/delete/execute
Purpose-based governance: Security can't be purely permission-based — must account for agent intent. Sales prep agent only needs read access to CRM, not delete database tables
Least privilege policy drift: When you understand what an agent is supposed to do, you can evaluate whether permissions match scope — and in practice they rarely do; gap widens over time
Continuous governance: Access reviews/audits are point-in-time checkboxes and false security. Agents change, instructions update, user bases shift, integrations expand — governance must be continuous to catch agents accessing applications outside normal pattern
AI agents create, use, and rotate identities at machine speed, outpacing traditional IAM controls

Enterprises that succeed won't block agents entirely — they'll make them governable and promote secure AI innovation
Treat agents as first-class identities with owners, access, behavior, risk, and lifecycle controls (Thiemann/Omdia)
AI agents are becoming privileged insiders; security and identity programs must catch up before those insiders become invisible attack paths
Need: consistent identity models, centralized enforcement, clear ownership, and continuous visibility (Gravitee)

Draft title "Agents as Identities: The IAM Gap Every Enterprise Is Ignoring" is an original rewrite — does not copy or closely mirror any source headline
Thiemann's TechTarget opinion piece provides strategic framework; our title focuses on the IAM governance gap
Gravitee headline: "Adoption Is Outpacing Control — The State of AI Agent Security 2026"

Dark Reading article (Todd Thiemann, Omdia principal analyst): Verified URL provided by user. Opinion piece on identity security for AI agents — one of the clearest frameworks available.
BleepingComputer article (sponsored by Token Security, June 2026): Verified via web_extract. Contains CSA 2026 survey stats (82%, 41%, 65%, 61%), identity sprawl narrative, continuous governance argument.
Gravitee State of AI Agent Security Report (April 2026, n=750): Verified via web_extract. Contains deployment scale data (doubling in 4 months), 48% unsecured, 54% incidents, six failure patterns, pre-deployment governance gaps.
CSA 2026 Survey: Referenced in BleepingComputer article as commissioned by Token Security. Stats: 82% discovered agents without security knowledge, 41% multiple times.
Gartner 2026 prediction: Cited in Gravitee report — Fortune 500 will have 150,000+ agents by 2028.

Primary: "Agent Sprawl and Shadow AI" (id: 9060b5cd-f42c-4727-b4a7-25e9d78cc52d) — domain: AI & Cybersecurity Governance
Secondary: "Access Management & IAM Security" (id: 3a7987b3-d077-48cb-a37f-45d127ea7a76) — domain: Security & Compliance

Research notes