Skip to content

Anoryx Research

Every company keeps a list of its employees. Almost none keep a list of everything else that can log in, spend money or read data on their behalf.

who-can-act / register

Actors that can log in

6 shown · 4 with no owner

  • Priya N.

    Employee · Finance

  • svc-billing-sync

    Service account

    no owner
  • invoice-agent

    AI agent

  • gh_pat_…7f2

    API token

    no owner
  • notes-ai

    Shadow AI app

    no owner
  • K. Iyer

    Left in August

    no owner

Definition

Identity debt is what builds up when a company does not know every actor that can act inside it, what each can reach, and who is accountable for it.

Like technical debt, it accumulates silently and is paid with interest during an incident. The credentials rarely do anything unusual. They simply still work.

Illustration: identity debt rises slowly over time, then spikes at an incident.Keys issued, apps adopted, people leaveIncident
Illustrative, not to scale Paid with interest

The access model

Five parts to every access decision.

An actor presents a credential to obtain a permission. Behind each actor there should be an owner who answers for it, and a lifecycle: created for a reason, used, reviewed, retired.

Actor

Who or what is acting?

Priya, accounts payable

Credential

What does it present?

SSO login with MFA

Permission

What can it do?

Approve invoices up to a limit

Behind every actor

Owner

Which human answers for it?

Her line manager, via HR

Lifecycle

Created, reviewed, retired?

Joiner → mover → leaver

Security programmes have perfected all five parts — for people.

One root cause, three symptoms

The same missing map, showing up in three places.

Each looks like a separate problem with a separate budget. Underneath, it is one question nobody can answer: what exactly do we have, and who owns it?

01

Machines with keys and no owner

API keys, OAuth tokens, service accounts and AI agents acting on a manager’s authority — created for a reason nobody wrote down.

A factory that hands spare keys to every contractor, delivery robot and cleaning crew, never writes down who has which key, and never changes the locks.
88%
of organisations define a ‘privileged user’ as human onlyCyberArk, 2025
53%
of open-source MCP servers rely on long-lived static keysAstrix Security, 2025
02

Work done outside the company’s sight

Shadow IT and shadow AI: tools adopted because the approved path is slow, now holding company data no one has mapped.

Workers who bring their own power tools and extension cords onto the shop floor because the approved ones take three weeks to order. Productive, until one sparks.
78%
of AI users bring their own AI tools to workMicrosoft, 2024
70%
of SaaS spend sits outside ITZylo, 2025
03

People who leave, access that stays

Offboarding closes the obvious doors. The supplier portal, the shared admin login and the tokens they created stay open.

An employee hands in their ID card on the last day but keeps the master key, the garage remote and the password to the supplier portal.
89%
of contractors kept some access after leavingSagiss 2026 via HRD Australia; indicative
22%
of breaches start with stolen credentials, the top initial access vectorVerizon DBIR, 2025

By the numbers

82–0

machine identities for every human identity

CyberArk, 2025 · Palo Alto Networks, 2026

0%

of former employees still had an account that worked

Sagiss 2026 via HRD Australia · indicative

0%

of secrets leaked on public GitHub in 2022 were still valid four years later

GitGuardian, 2026

+$0K

added to the average breach where shadow AI use is high

IBM, 2025

Figures are drawn from the sources named; several are vendor-sponsored surveys. Ranges are shown as ranges.

Why it is ignored

Five mechanisms. None of them is ignorance.

If the evidence is this strong, why does nearly every organisation carry identity debt? Because the incentives point the other way.

01

No line on the P&L

An orphaned token costs nothing until it is used. Budgets fund visible costs; identity debt is invisible by construction.

02

Diffuse ownership

DevOps creates service accounts, business units buy SaaS, HR owns the leaver event. A problem owned by everyone is managed by no one.

03

Definitions exclude machines

When ‘privileged user’ means a person, the largest population of privileged actors falls outside the programme.

04

Productivity bias

Shadow tools and quick integrations deliver real speed today. The risk is probabilistic and deferred.

05

The ‘believed unused’ fallacy

Every organisation has credentials it assumes are dead. Attackers only need one that is not.

What good looks like

Not conceptually complex. Operationally complex.

The controls are well understood. The hard part is building and maintaining a truthful inventory with owners — and keeping it true as people, tools and agents change every week.

01

Discovery

A continuously updated inventory of every actor: human, machine, agent and shadow app.

02

Ownership

Every actor linked to a named human who answers for it — and re-assigned when that person moves.

03

Least privilege

Permissions trimmed to what is actually used, not what was granted on day one.

04

Short-lived credentials

Tokens that expire by default, so a forgotten key stops working on its own.

05

Rotation and retirement

Keys rotated on schedule and actors retired when their purpose ends — including the leaver’s machine footprint.

06

Agent governance

Every AI agent registered with its delegating human, its scope and an expiry date.

The regulation clock

Two dates that turn a hygiene issue into a board issue.

Neither law says ‘identity debt’. Both expect you to show who and what can reach the data and systems you are responsible for.

India · DPDP Act

—days to go

Core duties become binding on 13 May 2027.

Penalties of up to ₹250 crore. Knowing who and what can reach personal data is where the evidence starts.

Source: SCC Online, 2025

Germany · NIS2

—days in force

Applies in Germany since 6 December 2025.

More than 30,000 entities are in scope. The directive lists access control among the risk-management measures it requires.

Source: Scheja & Partners, 2026

How Anoryx helps

Draw the map once. Then keep it true.

Start with a fixed-scope assessment that tells you where you stand. Keep the map current, inside your own infrastructure, with Sentinel.

Service · 3–4 weeks

Identity Debt Assessment

Read-only discovery, an ownership map, a risk register, a DPDP / NIS2 / ISO 27001 evidence gap list and a remediation roadmap. Nothing in your environment changes while we look.

How the assessment works
Product · self-hostedIn development

Sentinel

Find every key, bot, AI agent and shadow AI app. Give each one an accountable owner. Prove it to your auditor. An ownership control plane for non-human identities that runs inside your infrastructure.

Explore Sentinel

Figures on this page are drawn from the sources named; several are vendor-sponsored surveys. Anoryx produces audit-ready evidence that maps to these regulations; it does not certify compliance.

Find out who — or what — holds your keys. Before an incident tells you.

Start with a read-only Identity Debt Assessment. You keep the map, the register and the roadmap, whatever you decide next.