Use Case: AI Security
AI agent identity sprawl
A security audit found 340 active API keys and service accounts created for AI integrations over 18 months. 240 had no documented owner. 47 had permissions far beyond what their documented purpose required.
Every AI integration creates a new identity
The velocity of AI adoption in 2024-2025 has created an identity governance crisis. Each new AI workflow — a Slack bot, a code review agent, a customer service AI, a data pipeline assistant — creates one or more service accounts, API keys, and OAuth app registrations. These identities often have broad permissions because the developers want the AI to be capable, and they often lack human owners because they're seen as infrastructure rather than identities.
A compromised AI service account can be more dangerous than a compromised human account: it has no MFA, it operates continuously, its activity is hard to distinguish from normal AI behavior, and it often has access to data repositories, communication channels, and APIs that a human account wouldn't require.
detection: ai_identity_catalog_and_sprawl
type: identity_classification + governance_check
sources:
- aws_iam (service_accounts matching AI patterns)
- okta / entra_id (service_apps, OAuth_apps)
- github_apps (AI code tool registrations)
- api_gateway_logs (AI framework user agents)
classification:
- ai_identity: user_agent IN [langchain, openai, anthropic, llamaindex]
OR account_name MATCHES [ai_agent, llm_bot, copilot, gpt]
OR oauth_app.permissions INCLUDES llm_api_scopes
governance_check:
- has_owner: ai_identity.owner IN active_employees
- has_justification: created_with_ticket_or_approval
- least_privilege: actual_permissions vs used_permissions_90d
- active: last_used < 90d
flag: uncatalogued OR over_privileged OR no_owner OR staleMITRE coverage
AI identity sprawl FAQ
What is AI identity sprawl?
AI identity sprawl is the accumulation of uncatalogued service accounts, API keys, and OAuth applications created to power AI agents and integrations. As organizations adopt AI tools rapidly, each new AI workflow creates new identities with permissions — often without going through the normal identity governance process, leaving security teams without visibility into what AI identities exist and what they can access.
How does AI identity sprawl differ from traditional non-human identity sprawl?
AI service accounts accumulate permissions differently than traditional service accounts. AI agents often need broad access to be useful — read access to email, file systems, databases, APIs — making the principle of least privilege harder to enforce. Additionally, AI agents are often deployed without explicit security review because the developers see them as 'just software' rather than identities with security implications.
How does ManySignal discover AI-specific identities?
ManySignal classifies identities in the entity graph based on their authentication patterns. AI agents have distinct signatures: they authenticate via API key or service account token (not human credentials), they call APIs on regular schedules or in bursts, their user agent strings often contain LLM framework identifiers (LangChain, OpenAI, Anthropic), and their permissions often include LLM-specific API scopes.
What's the right governance process for AI agent identities?
ManySignal recommends the same governance process as other service accounts: formal provisioning with a named human owner, defined permission scope with a business justification, 90-day access review cycle, and automatic expiry for identities linked to time-limited projects. The AI identity catalog provides the inventory for this governance process.
Catalog every AI agent identity in your environment
AI-specific identity classification, permission audit, and governance workflow — built for the AI adoption wave.