Providers, models, agents and keys are an asset class of their own: procured in an afternoon, metered by the token, holding standing credentials and reading regulated data — with no owner, no cost centre and no retirement. This is where that gets governed.
Inventory tells you what you own. Intelligence tells you what it’s doing. The assets that matter most now have every property of an asset — and none of the controls.
A key was spun up in an afternoon. An agent was granted a scope. A vendor shipped a copilot into a tool you already paid for. A contractor left in March and the laptop came back, but the OAuth grant didn't. Each step was individually reasonable. None of them created a record.
Knowing where an asset sits tells you nothing about what it does. That gap is where the risk lives — and the evidence says it is already being exploited.
Signed up for in minutes, on a corporate card or a free tier, by someone who needed it that day.
Metered by the million tokens. For most finance teams the first governance signal is the invoice.
Standing keys and OAuth grants — a non-human identity with no MFA to challenge and no manager to notice.
Customer, contract, health and payment data — under terms that may permit training on what you send.
Forty pages of T&Cs, changed unilaterally, governing data ownership, retention and liability.
What your agent says, you said. A tribunal has already ruled on exactly that point.
No custodian, no cost centre, no risk class. Nobody is accountable when it behaves badly.
No renewal date, no end-of-life, no disposal certificate. It simply keeps running — and keeps spending.
of breached organisations reported a security incident involving shadow AI — more than double the year before.
IBM Cost of a Data Breach 2026 · 602 orgsaverage cost of a breach involving shadow AI, against a $4.99M global average.
IBM Cost of a Data Breach 2026of breached organisations had no AI governance policy — none at all, or still drafting one.
IBM Cost of a Data Breach 2026generative-AI data-policy violations per month at the average organisation; 2,100 in the top quartile.
Netskope Cloud & Threat Report 2026And because business heads now drive adoption, it isn't happening at the edges of the company. It is happening in every department at once — each on its own budget, its own timeline, its own vendors.
Dozens of ideas, no priorities
Every team running its own experiment, none of them compared.Tools nobody approved
Signed up on a card or a free tier, by someone who needed it that day.Data going places nobody tracks
Customer records and contracts pasted into models under terms nobody read.No one accountable when it breaks
No owner, no cost centre, no risk class — so no one to call.The alert fires.
A key appears with no owner and no risk class.Someone mentions it in Slack.
Everyone agrees it should be looked at.A workaround is agreed.
The pilot is too useful to switch off.The exception is the baseline.
Finance has re-forecast around it.Nobody panicked — an active exposure simply stopped looking like a threat and started looking like normal business. In AssetZentri, an exception gets an owner, an expiry and a record. Drift needs all three to be missing.
And it doesn't stay inside your walls. Your intelligence now runs on endpoints you don't own, through servers nobody inventoried, via agents holding standing credentials, inside vendors' infrastructure, under terms nobody re-read. The risk moved outside the perimeter. The register stayed inside it.
of generative-AI users at work still use personal, unmanaged AI accounts.
Netskope Cloud & Threat Report 2026of cloud environments now contain MCP — the protocol that lets agents reach your systems.
Wiz Research · July 2026of internet-exposed MCP servers return real data when called with no authentication at all.
Wiz Research · July 2026organisations had CRM data exported through one AI vendor's stolen OAuth tokens.
Salesloft Drift · August 2025Every domain answers two questions from the same record: what is it doing, and is it allowed to. Each has its own page.
Providers, models, agents and keys as tracked assets — judged at the call, logged as evidence.
Open the domain → DOMAIN 02Non-human identity with an owner, a scope and an expiry, and drift detected against the baseline.
Open the domain → DOMAIN 03Five-channel discovery, shadow-AI detection, renewal radar — and the forward AI meter.
Open the domain → DOMAIN 04Terms scored across eight risk categories, then enforced as rules at the model boundary.
Open the domain → DOMAIN 05One control evidencing many frameworks, collected as you operate rather than before the audit.
Open the domain → DOMAIN 06Hardware, software, cloud and SAM reconciled from fifteen sources into one authoritative record.
Open the domain →Five steps, in order. Legacy tools stop after the first and call themselves a platform. Nothing acts without your approval, and every action is written to an immutable record.
Illustrative of product behaviour, not customer data. Switch enforcement off to see what a measurement tool records: the same calls, every one allowed, and a report afterwards.
A signal no single tool sees — an unused licence, on a device past end-of-life, running a key whose owner left, under a vendor term that just changed.
Ranks it by cost and risk against your policy, then drafts the specific remediation: reclaim, revoke, rotate or renegotiate.
Nothing acts alone. You approve, defer or dismiss — with an impact preview before anything runs.
Scoped agents carry it out behind an in-boundary LLM, a sandbox and a one-click kill switch.
Every action recorded immutably and HMAC-signed, cross-mapped to every framework it satisfies. Evidence for the next audit, collected as you operate.
Your people are showing you exactly where AI could help. They just have no one to bring it to. Governance fixes that — not by saying no, but by making choices.
A governance policy nobody can enforce is a document. Here is what each of the four choices looks like once it is running.
Every model, agent and API key is registered with a named owner and an approval path. No owner, no access — the call is refused, not logged.
Providers approved before consumption. Five-channel discovery shows everything outside the list — so the list describes what's actually running, not what was intended.
Classification gates at the model boundary. Personal and regulated data is refused before it reaches a model that hasn't been cleared for it.
Consumption attributed per user, per project and per agent, with budgets that enforce themselves. A pilot ends with a number, not an impression.
The real question isn't "How do we stop people using AI?"It's "Who owns the decision of where we use it?"
Chaos doesn’t come from too much AI. It comes from nobody being in charge of it.
Traditional tools counted what you own. AssetZentri governs what it does — one record, from laptops to LLMs. Connect the identity providers and MDM you already run. Discovery begins immediately, the registry builds itself, and ZentriPulse ranks what to fix first. The fastest way to understand AssetZentri is to watch it read an estate it has never seen.
Inventory is a solved problem — solved for an estate that sat still. Fifteen discovery sources merge into one authoritative record for hardware, software, SaaS, cloud, identities and contracts. Then AI reads all of it and ranks what actually matters.
sk-prod-4471 was issued by the Design team, in an afternoon, for a pilot. Here is what five separate tools know about it — and what one record knows.
Contractor offboarded on 14 March. Account disabled, laptop returned.
✓ Offboarding completeWS-4471 is past end-of-life. Three Adobe seats still assigned to it.
Scheduled for refresh41,280 tokens this month on an API key in the company account.
Within account limitTraffic to a model endpoint that has no data classification.
No policy matchThe model provider disclosed a security incident last week.
Noted
sk-prod-4471INCIDENT · 5 CORRELATED FINDINGS
Owner: none · issued by the Design team · 90 days old
Five tools each saw one harmless fact. One record sees an incident — and one fix serves every stakeholder at once.
Illustrative reference-tenant scenario.
Three questions, asked of fifteen systems that have never been asked them together. Reconciling the answers is the part that has to happen before any of the six domains can act.
Source types, not a certified connector list. Coverage depends on what your estate runs and what each vendor exposes — a walkthrough confirms which of these we can read on day one.
Reconciling fifteen sources produces one record per asset. It also produces more findings than any team can work through. ZentriPulse is what turns the record into an order of work.
A queue of alerts is a list of things nobody ordered. ZentriPulse reads the whole record — exposure, ownership, contract, spend — and puts the estate in the order it should be worked.
Every action arrives as a proposal with an impact preview: what changes, what breaks, who is affected. Nothing executes on its own.
The rank, the inputs and the decision are written to the same append-only record as everything else — so the next person can see why it was first.
Every domain answers two questions from the same record: what is it doing, and is it allowed to. Each has its own page.
Providers, models, agents and keys as tracked assets — judged at the call, logged as evidence.
Open the domain → DOMAIN 02Non-human identity with an owner, a scope and an expiry, and drift detected against the baseline.
Open the domain → DOMAIN 03Five-channel discovery, shadow-AI detection, renewal radar — and the forward AI meter.
Open the domain → DOMAIN 04Terms scored across eight risk categories, then enforced as rules at the model boundary.
Open the domain → DOMAIN 05One control evidencing many frameworks, collected as you operate rather than before the audit.
Open the domain → DOMAIN 06Hardware, software, cloud and SAM reconciled from fifteen sources into one authoritative record.
Open the domain →Inventory is a solved problem — solved for an estate that sat still. Seven rungs sit between a list and a decision. Six vendors own the first six, and each owns a different record. Select any rung.
Buying five of these doesn't add up to one of them — the facts still live in five places.
| Capability | ITAM & measurement tools | AssetZentri |
|---|---|---|
| What it does with an asset | Counts it | Judges it |
| Output | A list, refreshed nightly | A verdict, at runtime |
| A call over budget | Reported afterwards | Refused before it goes out |
| An agent tool-call | Invisible, or observed only | Allow · deny · throttle · require approval |
| Non-human identity | Not modelled | Registered, owned, bound to credentials |
| Data classification | Out of scope | PII gated at the model boundary |
| Audit evidence | Assembled for the audit | Collects itself, HMAC-signed |
| Scope | Hardware & licences, or AI only | Laptops to LLMs — one record |
Scroll the table sideways to compare →
Gross savings are a brochure. Risk-adjusted savings are a business case — every benefit weighted by how confident we actually are, and the one exposure we refuse to put a number on. And none of it is a new budget line: it is waste already on the invoice.
Illustrative. Proportions and confidence weights come from a reference tenant with $2.09M annual SaaS and licence spend. Your estate will differ — connecting billing, GL and contracts replaces these with your own numbers.
| Lever | Gross | Confidence | Risk-adjusted |
|---|
It compounds, too. Revoking one stale admin grant is simultaneously evidence for SOC 2, ISO 27001, SEBI and DPDP. Reclaiming an idle seat is a cost saving, a licence position and SAM evidence at once. Governance stops being a tax the moment one action counts everywhere it applies.
The seats are still idle, the duplicates still billed, the over-provisioned plans still renewing at full tier.
Renewing on terms nobody re-read — and the leverage window closed months before the date.
Full compliance lands on 13 May 2027 with no grace period. Evidence you haven’t started collecting cannot be backdated.
Under-licensing sits on the estate, waiting for a vendor audit to find it first.
sk-prod-4471 — still ownerless, still spending, still reading data it was never cleared for.
Same estate, same blind spots, same four numbers that don’t agree.
Doing nothing isn’t the free option. It’s the one whose price never shows up on an invoice.
Figures scale with the spend entered above; counts are from the reference tenant.
Traditional tools counted what you own. AssetZentri governs what it does — one record, from laptops to LLMs. Connect the identity providers and MDM you already run. Discovery begins immediately, the registry builds itself, and ZentriPulse ranks what to fix first. The fastest way to understand AssetZentri is to watch it read an estate it has never seen.
Controls map themselves to assets, policies and configurations — cross-mapped across 19 frameworks and refreshed nightly. Evidence accumulates as you operate rather than being rebuilt each quarter.
Most AI-governance marketing points at Brussels. It shouldn't — the EU's high-risk regime was deferred in July 2026 to December 2027 and August 2028. The nearest binding dates for an Indian enterprise are much closer, and they read almost line for line like asset governance.
Encryption, obfuscation, masking or virtual tokens. Access control. Logs, monitoring and review to detect unauthorised access. Backups. And retention of logs for a minimum of one year — with equivalent safeguards contractually flowed down to every processor. Failure here carries the highest penalty in the Act.
Without delay, to affected individuals and to the Board. Then a detailed follow-up to the Board within 72 hours — causes, mitigation, who caused it, remedial measures. Calling this simply "72 hours" understates it, and you cannot assemble that second report from systems that were never logging.
An annual DPIA. An annual audit whose significant findings go to the Board. An India-based DPO. And the provision that should concern every AI programme:
…observe due diligence to verify that technical measures, including algorithmic software deployed for … processing of personal data, are not likely to pose a risk to the rights of Data Principals.
Stated precisely: SDF duties bind only entities notified as Significant Data Fiduciaries, and none have been notified yet. The Data Protection Board is constituted but not yet staffed. DPDP is not a blanket localisation regime.
The honest position on the EU: as of September 2026 there is no published fine under the AI Act. Prohibitions, AI literacy and Article 50 transparency are live; the high-risk regime was moved to 2 December 2027 and 2 August 2028 by Regulation (EU) 2026/1744. Liability for AI failures is arriving today through consumer and data-protection law instead.
Meanwhile SEBI CSCRF deadlines have already passed — it mandates an asset inventory with criticality classification and an SBOM for critical systems. The RBI issued seven new cyber and technology-risk Directions on 31 July 2026, effective immediately, also requiring an information asset inventory with criticality classification. DORA has required an ICT asset inventory and a register of third-party arrangements since January 2025. Four regulators, asking for the same artifact.
A control does not belong to a standard — it belongs to an asset. Map it once to the asset and the same evidence answers several auditors at once.
Trust services criteria turn on who had access to what, and whether someone owned it. Control ownership and access evidence accumulate as you operate rather than being assembled in the weeks before fieldwork.
Annex A expects an asset inventory that is actually maintained, and the 2022 revision adds threat intelligence and cloud services. One record per asset, reconciled from fifteen sources, is what the statement of applicability is written against.
The AI management system standard asks for a model inventory with ownership and lifecycle. Providers, models, agents and keys are already tracked as assets here — with an owner, a purpose and a retirement date.
India’s data protection regime. The DPDP Rules 2025 (G.S.R. 846(E)) were notified on 13 November 2025, with full compliance due 13 May 2027. The question it forces is which model touched which personal data, under whose authority.
The Cybersecurity and Cyber Resilience Framework asks regulated entities for measurable control posture across the estate — not a point-in-time report, which is why evidence has to collect itself.
The IT governance and outsourcing directions already expect a defensible asset and access position from banks and NBFCs. AI arrived without one — holding standing credentials no joiner-mover-leaver process ever saw.
Information and cybersecurity guidelines put the burden on demonstrable control ownership for every system touching policyholder data — including the ones adopted faster than they were reviewed.
Obligations scale with the risk tier of each system, which presumes you can say what each system is for. A model inventory with a stated purpose per deployment is the precondition.
Cross-mapped across nineteen frameworks in total; the eight above are the ones we are asked about most, and a walkthrough confirms the rest against your own scope. Regulatory dates current as of 25 September 2026. Confirm exact clause mapping with your auditor — this is not legal advice.
In regulated entities the same blind spot carries a named supervisor and a date. The obligation is already written — what is missing is the record that answers it.
Outsourcing and IT governance directions already expect a defensible asset and access position. AI arrived without one.
The Cybersecurity and Cyber Resilience Framework asks for measurable control posture across the estate, not a point-in-time report.
Information and cybersecurity guidelines put the burden on demonstrable control ownership — for every system touching policyholder data.
Patient data under the DPDP Act, processed by tools that were adopted faster than they were reviewed.
Vendor terms that permit training on submitted data are an intellectual-property question long before they are a privacy one.
Data residency and in-boundary processing are procurement conditions, not preferences.
Regulatory context, not legal advice. Confirm applicability and timelines with your own counsel.
What the domain covers, what separates it from a GRC platform, and how one control ends up evidencing five frameworks at once.
"Trust us, it's in Virginia" is not an audit answer in Mumbai. The platform is built to the controls it audits you against.
Contracts and customer data analysed in-boundary. Nothing regulated leaves your residency.
A unique X.509 certificate per device, verified bidirectionally with mTLS on every request. No shared secrets, no reusable tokens.
An impact preview before anything runs, and scoped least-privilege permissions on every agent.
By construction, not by policy. Nothing executes unapproved — the approval is the control, not a setting.
Terminates SSO sessions and revokes across every IdP, OAuth grant, SaaS app and AI key in parallel — in minutes, not hours.
HMAC-signed and hash-chained, with before-and-after values on every change. Arguments are stored as SHA-256 digests only.
Traditional tools counted what you own. AssetZentri governs what it does — one record, from laptops to LLMs. Connect the identity providers and MDM you already run. Discovery begins immediately, the registry builds itself, and ZentriPulse ranks what to fix first. The fastest way to understand AssetZentri is to watch it read an estate it has never seen.
Providers, models, agents and keys are an asset class of their own — procured in an afternoon, metered by the token, holding standing credentials and reading regulated data. This is where each call is decided.
An agent crossed 162% of its token budget before anyone looked. Nobody capped it, because nobody owned it. Procurement assumes a signature — AI arrived without one, holding standing credentials and reading regulated data.
Procurement assumes a signature — an owner, a renewal date, a line in the register. Nothing about AI arrived that way. It spends metered budget, holds standing credentials and reads regulated data. Providers, models, agents and keys become tracked assets, with budgets that enforce themselves before the call goes out.
An AI observability tool tells you what a model did. A gateway can cap a key. Neither knows who owns the agent, whether the data it touched was classified, or what the contract permits — so neither can say no.
Model, prompt, classification and budget checked before the request leaves. Allow, throttle, approve or deny.
Agents run sandboxed with an impact preview first. Nothing reaches production unapproved.
Every verdict signed and immutable, with a kill switch behind it.
A log tells you what happened. A ledger tells you what was decided — and on whose authority.
LLM observability records calls. Gateways cap keys. Neither carries the owner, the classification or the contract clause. Only the asset record does.
The same domain answers to both pillars. AI Governance decides what runs; Asset Intelligence knows what you hold.
Model, prompt, data classification and budget checked at the boundary. Allowed, throttled or refused — and the verdict written down either way.
Four units of AI consumption normalised to one currency figure and attributed to a person and a project — not a workspace, not a key.
Same record, two readings. Neither works without the other — a policy needs an object to act on, and a record without policy can only report.
Every domain answers two questions from the same record: what is it doing, and is it allowed to. Each has its own page.
Providers, models, agents and keys as tracked assets — judged at the call, logged as evidence.
You are here → DOMAIN 02Non-human identity with an owner, a scope and an expiry, and drift detected against the baseline.
Open the domain → DOMAIN 03Five-channel discovery, shadow-AI detection, renewal radar — and the forward AI meter.
Open the domain → DOMAIN 04Terms scored across eight risk categories, then enforced as rules at the model boundary.
Open the domain → DOMAIN 05One control evidencing many frameworks, collected as you operate rather than before the audit.
Open the domain → DOMAIN 06Hardware, software, cloud and SAM reconciled from fifteen sources into one authoritative record.
Open the domain →Traditional tools counted what you own. AssetZentri governs what it does — one record, from laptops to LLMs. Connect the identity providers and MDM you already run. Discovery begins immediately, the registry builds itself, and ZentriPulse ranks what to fix first. The fastest way to understand AssetZentri is to watch it read an estate it has never seen.
Hardware, software, cloud and licence position reconciled into one authoritative record per asset — then ranked by what actually matters.
Fifteen systems each logged one harmless fact. None connected them. A register can count; it takes AI to judge — and to tell you before morning.
Inventory is a solved problem — solved for an estate that sat still. Fifteen discovery sources, each certain and none agreeing, merge into one authoritative record per asset, covering hardware, software, cloud and AI alike. Then AI reads all six lenses continuously and ranks what actually matters.
Discovery finds assets. A CMDB stores them. Neither reconciles fifteen certain, disagreeing sources into a record the other five domains can act on — which is why the second and third outcome usually cost as much as the first.
The same asset seen six ways, not six copies drifting apart.
Reclaim one idle seat and you have a cost saving, a licence position and SAM evidence from a single action.
AI reads all six lenses continuously and ranks what actually matters — the part no inventory has ever done.
Governance stops being a tax the moment it compounds.
Discovery tools find assets. CMDBs store them. Neither reconciles fifteen disagreeing sources into one record the rest of the estate can act on.
The same domain answers to both pillars. AI Governance decides what runs; Asset Intelligence knows what you hold.
Policy needs an object to act on. The register is what turns a rule into an enforceable decision — which is why governance without one stays advisory.
Each system certain and none agreeing. The record is what remains once they are reconciled — and AI is what ranks what matters in it.
Same record, two readings. Neither works without the other — a policy needs an object to act on, and a record without policy can only report.
Every domain answers two questions from the same record: what is it doing, and is it allowed to. Each has its own page.
Providers, models, agents and keys as tracked assets — judged at the call, logged as evidence.
Open the domain → DOMAIN 02Non-human identity with an owner, a scope and an expiry, and drift detected against the baseline.
Open the domain → DOMAIN 03Five-channel discovery, shadow-AI detection, renewal radar — and the forward AI meter.
Open the domain → DOMAIN 04Terms scored across eight risk categories, then enforced as rules at the model boundary.
Open the domain → DOMAIN 05One control evidencing many frameworks, collected as you operate rather than before the audit.
Open the domain → DOMAIN 06Hardware, software, cloud and SAM reconciled from fifteen sources into one authoritative record.
You are here →Traditional tools counted what you own. AssetZentri governs what it does — one record, from laptops to LLMs. Connect the identity providers and MDM you already run. Discovery begins immediately, the registry builds itself, and ZentriPulse ranks what to fix first. The fastest way to understand AssetZentri is to watch it read an estate it has never seen.
Controls map themselves to assets, policies and configurations across nineteen frameworks, refreshed nightly. Proof accumulates instead of being rebuilt.
A model read customer records it was never cleared for. When the auditor asks which model touched what, under whose authority, the answer has to exist already.
Under whose authority, from which device, against which control — and where the record is. Evidence gathered for an audit is stale the day after it’s gathered. Controls map themselves to assets, policies and configurations, cross-mapped across 19 frameworks and refreshed nightly, so proof accumulates as you operate.
A GRC tool is a place to put evidence you gathered somewhere else. When the evidence already lives on the asset record, the control is proved where the work happened rather than reconstructed before the audit.
Disk encryption from the agent, an access review from identity, a token ledger from AI — each already on the record.
A single access-review certification evidences SOC 2, ISO 27001, SEBI CSCRF and DPDP at once.
Posture scored continuously, so the anomaly is closed before it becomes the standard.
One certification. Five frameworks. Zero re-collection.
GRC platforms hold evidence. Scanners produce findings. Neither owns the record the evidence comes from — which is why the gathering never stops. Framework mappings are illustrative; validate exact clause mapping with your auditor before relying on it.
The same domain answers to both pillars. AI Governance decides what runs; Asset Intelligence knows what you hold.
Every decision leaves a record mapped to the framework that asked for it. The audit stops being a project and becomes a query.
One record read against nineteen frameworks at once, so the answer to “are we compliant” carries a date instead of a project plan.
Same record, two readings. Neither works without the other — a policy needs an object to act on, and a record without policy can only report.
Every domain answers two questions from the same record: what is it doing, and is it allowed to. Each has its own page.
Providers, models, agents and keys as tracked assets — judged at the call, logged as evidence.
Open the domain → DOMAIN 02Non-human identity with an owner, a scope and an expiry, and drift detected against the baseline.
Open the domain → DOMAIN 03Five-channel discovery, shadow-AI detection, renewal radar — and the forward AI meter.
Open the domain → DOMAIN 04Terms scored across eight risk categories, then enforced as rules at the model boundary.
Open the domain → DOMAIN 05One control evidencing many frameworks, collected as you operate rather than before the audit.
You are here → DOMAIN 06Hardware, software, cloud and SAM reconciled from fifteen sources into one authoritative record.
Open the domain →Traditional tools counted what you own. AssetZentri governs what it does — one record, from laptops to LLMs. Connect the identity providers and MDM you already run. Discovery begins immediately, the registry builds itself, and ZentriPulse ranks what to fix first. The fastest way to understand AssetZentri is to watch it read an estate it has never seen.
Service accounts, API keys, OAuth grants and agent credentials outnumber your people — and outlive them. Each one gets a named owner, a scope and an expiry.
The contractor left in March. The OAuth grant, the API key and the service account their agent runs on did not — no manager to notice, no MFA to challenge, no leaving date.
Offboarding is a checklist written for humans. The OAuth grant, the API key and the service account your agents run on have no manager to notice, no MFA to challenge and no leaving date. Full identity governance without a separate suite — and one switch that revokes across every IdP, SaaS app and AI key at once.
Your identity provider is excellent at humans. It has no concept of a service account that belongs to an agent, a key issued from a team inbox, or an OAuth grant with no leaving date.
Service accounts, API keys and agent credentials as tracked assets, each with a named custodian.
Issued, scoped, rotated, expired. Deprovisioning a human takes their machine identities with them.
Continuous comparison with the scope the credential started with, so a temporary elevation cannot quietly stay.
One revoked grant is four frameworks evidenced — from a single action.
IAM governs people. PAM governs privileged sessions. CIEM governs cloud entitlements. None of them ties a key to a contract, a budget and a data classification on one record.
The same domain answers to both pillars. AI Governance decides what runs; Asset Intelligence knows what you hold.
Non-human identities get what people get: a named custodian, a scope, a review date. When the owner is deprovisioned, the credential goes with them.
Service accounts, OAuth grants and agent credentials reconciled against the directory, so a non-human identity always resolves to someone accountable.
Same record, two readings. Neither works without the other — a policy needs an object to act on, and a record without policy can only report.
Every domain answers two questions from the same record: what is it doing, and is it allowed to. Each has its own page.
Providers, models, agents and keys as tracked assets — judged at the call, logged as evidence.
Open the domain → DOMAIN 02Non-human identity with an owner, a scope and an expiry, and drift detected against the baseline.
You are here → DOMAIN 03Five-channel discovery, shadow-AI detection, renewal radar — and the forward AI meter.
Open the domain → DOMAIN 04Terms scored across eight risk categories, then enforced as rules at the model boundary.
Open the domain → DOMAIN 05One control evidencing many frameworks, collected as you operate rather than before the audit.
Open the domain → DOMAIN 06Hardware, software, cloud and SAM reconciled from fifteen sources into one authoritative record.
Open the domain →Traditional tools counted what you own. AssetZentri governs what it does — one record, from laptops to LLMs. Connect the identity providers and MDM you already run. Discovery begins immediately, the registry builds itself, and ZentriPulse ranks what to fix first. The fastest way to understand AssetZentri is to watch it read an estate it has never seen.
Five discovery channels find what single sign-on never saw, and consumption is priced forward — while there is still a month left to act on it.
A new AI tool was authorised through a personal account. Single sign-on records the apps that came through the front door; five discovery channels find everything that didn’t.
It faithfully records the applications that came through the front door. Five native discovery channels — email, browser, OAuth, IdP and network — find everything that didn’t, including the AI tools nobody approved. Then idle seats and renewals are priced before the auto-renewal makes the decision for you.
It is metered by the token, the GPU hour and the API call — and for most teams the first governance signal is the invoice, ninety days after the decision that caused it.
On day nine, the projection for day thirty, against a ceiling that can throttle or refuse. Not a report on last month.
Cost per query, per user, per resolved ticket — so someone can answer whether the copilot is worth it.
Five-channel discovery finds the personal-account and card-paid AI, then folds it into the same budget line.
Know the bill on day nine, not day thirty.
FinOps prices cloud after the invoice. Gateways meter keys without context. ITAM knows the contract and has never metered a token. Only the asset record holds all three.
The same domain answers to both pillars. AI Governance decides what runs; Asset Intelligence knows what you hold.
New AI features arrive inside tools you already bought, on terms you already signed. Discovery finds them; policy decides whether they stay on.
Seats, renewals and AI consumption on one meter, with a forward run-rate against the budget — while there is still a month left to act on it.
Same record, two readings. Neither works without the other — a policy needs an object to act on, and a record without policy can only report.
Every domain answers two questions from the same record: what is it doing, and is it allowed to. Each has its own page.
Providers, models, agents and keys as tracked assets — judged at the call, logged as evidence.
Open the domain → DOMAIN 02Non-human identity with an owner, a scope and an expiry, and drift detected against the baseline.
Open the domain → DOMAIN 03Five-channel discovery, shadow-AI detection, renewal radar — and the forward AI meter.
You are here → DOMAIN 04Terms scored across eight risk categories, then enforced as rules at the model boundary.
Open the domain → DOMAIN 05One control evidencing many frameworks, collected as you operate rather than before the audit.
Open the domain → DOMAIN 06Hardware, software, cloud and SAM reconciled from fifteen sources into one authoritative record.
Open the domain →Traditional tools counted what you own. AssetZentri governs what it does — one record, from laptops to LLMs. Connect the identity providers and MDM you already run. Discovery begins immediately, the registry builds itself, and ZentriPulse ranks what to fix first. The fastest way to understand AssetZentri is to watch it read an estate it has never seen.
Training rights, retention, liability and breach history, scored from the vendor’s own terms — and turned into rules that hold at the model boundary.
A contract auto-renewed on forty pages of terms nobody re-read — including the clause that lets the vendor’s model learn from what you send it.
Forty pages of terms, changed unilaterally, governing who owns your data, how long it is kept and what a model may learn from it. Paste the vendor’s terms URL — no document upload. AI scores the terms across eight risk categories, and the Policy Comparison Engine flags every clause that conflicts with your own.
Contract lifecycle tools file the agreement and tell you when it expires. Third-party risk scores the vendor. Neither turns a training-rights clause into a rule that stops the call which would breach it.
Forty pages scored across eight risk categories from the vendor’s own terms URL — no document upload.
A training-rights clause turns into a data-classification rule enforced at the model boundary.
The notice window, the seat utilisation and the re-scored terms arrive together, while there is still room to negotiate.
The leverage window closes months before the renewal date.
CLM stores the contract. Procurement suites track the spend. TPRM scores the vendor. None of them turns a clause into a runtime rule.
The same domain answers to both pillars. AI Governance decides what runs; Asset Intelligence knows what you hold.
Clauses on training rights, retention and liability are read once and enforced continuously — so the agreement governs the model instead of describing it.
Every agreement carries its notice window, its seat utilisation and its re-scored terms, surfaced before the window closes rather than after.
Same record, two readings. Neither works without the other — a policy needs an object to act on, and a record without policy can only report.
Every domain answers two questions from the same record: what is it doing, and is it allowed to. Each has its own page.
Providers, models, agents and keys as tracked assets — judged at the call, logged as evidence.
Open the domain → DOMAIN 02Non-human identity with an owner, a scope and an expiry, and drift detected against the baseline.
Open the domain → DOMAIN 03Five-channel discovery, shadow-AI detection, renewal radar — and the forward AI meter.
Open the domain → DOMAIN 04Terms scored across eight risk categories, then enforced as rules at the model boundary.
You are here → DOMAIN 05One control evidencing many frameworks, collected as you operate rather than before the audit.
Open the domain → DOMAIN 06Hardware, software, cloud and SAM reconciled from fifteen sources into one authoritative record.
Open the domain →Traditional tools counted what you own. AssetZentri governs what it does — one record, from laptops to LLMs. Connect the identity providers and MDM you already run. Discovery begins immediately, the registry builds itself, and ZentriPulse ranks what to fix first. The fastest way to understand AssetZentri is to watch it read an estate it has never seen.
Public incidents, the recurring scenarios that never get reported, and what we are watching next — each mapped to the governance gap that allowed it and the control that would have closed it. Updated as new events surface.
Each follows the same shape: an asset nobody had on a register, doing something nobody had approved, until someone outside the company reported it. Filter by the domain it belongs to.
Summarised from public reporting and primary disclosures, linked on each card. Described to illustrate a governance gap, not to attribute fault.
These aren't incidents anyone reported. They are the recurring shapes underneath the ones who do — each starting as something reasonable, and each ending as an exception nobody closed.
Composite patterns, not customer accounts. Described to show a governance gap that recurs across estates.
Four open questions that will move this page. Each says where things actually stand today — including where the honest answer is “nothing has happened yet.”
Rules notified 13 November 2025. Consent Manager registration provisions take effect 13 November 2026, and full substantive compliance lands 13 May 2027 with no grace period. The Data Protection Board is constituted but not yet staffed, and has issued no orders.
Watching: the first Significant Data Fiduciary notifications, and whether the January 2026 acceleration proposal is revived.
Article 50 transparency and the prohibitions are live. The high-risk regime was deferred by Regulation (EU) 2026/1744 to 2 December 2027 for Annex III and 2 August 2028 for Annex I. As of today there is still no published fine under the Act.
Watching: the first enforcement action, and any figure that can finally be sourced.
MCP now appears in roughly 80% of cloud environments; about one in six exposes a server to the open internet, and 42% of those return real data to unauthenticated callers. Most still negotiate a protocol version that predates authentication. The NSA published its first MCP security guidance in June 2026.
Watching: whether authenticated versions are actually adopted, or simply shipped.
A malicious package impersonating official tooling silently copied every email an agent sent, and an over-scoped CI token put a destructive prompt into a coding assistant used by around a million developers. Both arrived through ordinary dependency updates, not through an attack on the enterprise.
Watching: whether agent registries and skill marketplaces add meaningful review before distribution.
How this page is maintained. Reviewed monthly; last reviewed 23 September 2026. We list only what we can source, and we link the source on every card — if a claim can’t be traced to a primary disclosure or credible reporting, it isn’t here. Several widely-circulated figures in this field fail that test, including more than one confident number about AI Act penalties.
Traditional tools counted what you own. AssetZentri governs what it does — one record, from laptops to LLMs. Connect the identity providers and MDM you already run. Discovery begins immediately, the registry builds itself, and ZentriPulse ranks what to fix first. The fastest way to understand AssetZentri is to watch it read an estate it has never seen.
AssetZentri answers them all.
Governance infrastructure for the next-generation enterprise.
Your smartest asset isn’t on the register. Traditional IT got intelligent — the laptop runs a copilot, the service account belongs to a bot. Traditional ITAM still only counts. Nobody owns the asset that thinks.
AssetZentri registers what thinks. Every model, agent and key on one record — tokens, GPU hours, API calls and cloud metered to a user and a project, every call judged against policy, every action left as evidence.
Decide what runs, know what you hold — and let the record say which of it to fix first. Everything else on this page is one of these three, in detail.
Six things happen in sequence — something spends, something signs, something lets a stranger in — and not one of them is asked to explain itself. Watch the two minutes, or read the six.
AssetZentri campaign film · 2:04. Nothing loads until you press play.
Jump to any chapter, or read it instead — start with AI Governance.
Nobody saw it because nobody was built to. For twenty years the question was custodial: how many laptops, are they patched, who has the licence. An inventory answers that.
The questions arriving now are about risk, cost and compliance at once — and they come from the business. Neither business IT nor traditional IT can answer them alone.
“What does it risk, what will it cost — and can we prove we’re allowed to run it?”
Risk, cost and compliance stopped being three conversations the moment all three depended on the same record. Governance is what answers the question — and governance is exactly what neither side was built to do.
“Which model read the customer table today — and was it allowed to?”
0 results
It knows the asset exists, who bought it and where it sits. It was never built to know what the asset did.
gpt-4o · 3 calls · 14:02–14:06 owner: analytics-svc · project: CX-forecast policy: allowed · classification: PII · logged as evidence
Same question, answered from the record — and left behind as proof you can hand an auditor.
Illustrative of the query, not a customer account.
The register makes AI governable. AI makes the register intelligent. Everyone else is holding one end.
Which is why the answer has two halves. AI Governance puts an owner, a budget and a policy on everything that thinks. Asset Intelligence gives it a record to be judged against.
The invoice isn’t a governance signal — it’s a post-mortem. You cannot budget what nobody is metering.
Tokens, GPU hours, API calls and cloud, normalised to a single figure.
Attributed to a person and a project — not a workspace, not a key.
Run-rate projected to period end, against a ceiling that can refuse.
The invoice arrives on the 30th. The overrun happened on the 9th.
FinOps prices cloud after the invoice. Gateways meter keys without context. ITAM knows the contract and has never metered a token. Only the asset record holds all three.
A control belongs to an asset, not to a standard. Map it once and the same evidence answers several auditors at once.
Public, dated and sourced. Each follows the same shape: an asset nobody had on a register, doing something nobody approved. Every card links to the primary source.
Answered plainly, including the one about what we can actually prove.
The same record, read from wherever you sit — and the two pages most people open first. Start anywhere; it is the same estate.
You can name every asset you own.
Now name one thing they did today.
That second list is the one your auditor, your CFO and your attacker are all working from. AssetZentri is where it lives.
Traditional tools counted what you own. AssetZentri governs what it does — one record, from laptops to LLMs. Connect the identity providers and MDM you already run. Discovery begins immediately, the registry builds itself, and ZentriPulse ranks what to fix first. The fastest way to understand AssetZentri is to watch it read an estate it has never seen.
The CTO’s asset count doesn’t match Finance’s spend. GRC’s evidence doesn’t match either. Business IT isn’t in any of them. It was never a data problem — it’s an architecture problem.
You inherit everyone else’s decisions. The CTO’s asset count doesn’t match Finance’s spend. GRC’s evidence doesn’t match either. Business IT isn’t in any of them. It was never a data problem — it’s an architecture problem.
Whether IT has been narrowed to governance or pulled back to the centre, the requirement converges: know what is running, what it touched, and who decided.
For the CIO and Head of IT whose job has become governance — and for the business heads choosing the tools.
For banks, NBFCs, insurers and market intermediaries — the CIO, CISO, CRO and Compliance teams who answer to a regulator.
Shadow AI is sharpest in small and mid-sized companies, where there is no dedicated governance team at all. Row-level tenant isolation, automated provisioning and per-tenant configuration let an MSP govern every client from one console — with each client’s data and evidence kept separate.
The whole estate on one number the board can be given. Lifecycle from procurement to certified disposal, licence position proved rather than estimated, shadow IT and end-of-life exposure surfaced early.
Three cost curves — legacy, cloud and AI — on one meter, attributed to a user and a project rather than a cost centre and a guess. Waste pre-quantified on the home screen, and a forward number before the spend.
Evidence collected as you operate and cross-mapped across 19 frameworks. Every standing credential, every non-human identity, every access review — with the audit becoming a download rather than a quarter.
Ask the estate in plain English — no report builder, no SQL. Approve a model, cap a budget, scope an agent, and get a team moving in an afternoon instead of a procurement cycle.
For twenty years, IT decided what the company ran. That era is over. Marketing chooses its AI, sales its copilots, HR its screening tools, finance its forecasting models — each funded from a business budget, each adopted on a business timeline.
IT's sphere of influence hasn't disappeared. It has narrowed to the one thing a business head can't do for themselves: govern the risk.
In regulated entities, the direction reverses. RBI, SEBI, IRDAI and DPDP make accountability non-delegable, so Enterprise IT is being pulled back to the centre.
Two opposite motions. Both need the same thing: one governed record of what is running, what it touched — and who decided.
Traditional tools counted what you own. AssetZentri governs what it does — one record, from laptops to LLMs. Connect the identity providers and MDM you already run. Discovery begins immediately, the registry builds itself, and ZentriPulse ranks what to fix first. The fastest way to understand AssetZentri is to watch it read an estate it has never seen.