Home / AI-SPM

Know every AI you run, what it can reach, and what to fix first.

AI security posture management is the practice of continuously finding the AI a company runs, working out how exposed it is, and governing it. Sargha does it from the endpoint to the cloud, read-only.

Your estate

Endpoints and serversSensor, read-only. Masks keys on the machine.

Agents
Tool servers
Keys

Cloud accountsAWS, Azure, Google Cloud through a read-only role

Identity and devicesOwners from your directory

SaaS and codeAI apps, libraries and pipelines

Sargha cloudYour tenant

CollectReceives over an encrypted connection and keeps your tenant apart.

One inventoryEvery AI found, de-duplicated, each with an owner.

Apps and agents
Tool servers
Keys
Identities

What each can reachFiles, credentials, cloud accounts and data, from the endpoint into the cloud.

Score with proofLikelihood times impact. Each fact shows its proof and what would remove it.

Decide and evidenceAn owner, an approved or not-allowed decision, a review date, signed reports.

Your tools and people

SIEMSplunk, QRadar, Sentinel, Elastic, Chronicle

TicketsJira, ServiceNow: one per risk, to its owner

ChatSlack, Microsoft Teams

AuditorsAI Bill of Materials and signed evidence

Metadata
Findings and alerts
Identity and access
Evidence and reports

Dashed: optional
Thicker: more data
Encrypted in transit
Read-only

1
AI found on machinesWhich AI tools, agents, tool servers and keys are present, with machine and user.Never sent Key values, file contents, prompts and responses, keystrokes.
2
Cloud accountsAI services, roles and storage that AI can use.
3
Identity and devicesOwners and accounts from your directory.
4
SaaS and codeAI apps, libraries and pipelines in use.
5
One inventoryEverything found, de-duplicated, with where and when it was seen.
6
ReachWhat each item can touch, and the paths between the endpoint and the cloud.
7
ExposureFacts about each item, scored by likelihood and impact.
8
DecisionA ranked list with an owner, a decision and a review date for each item.
9
To your SIEMFindings with context.
10
To ticketsOne ticket per risk, assigned to its owner.
11
To chatAlerts in the channel you choose.
12
EvidenceA signed report and an AI Bill of Materials, mapped to the frameworks you report against.

Discovery is read-only. Mapping and evidence are not a certification.

Starts where the AI actually is

A sensor that only reads by default covers Windows, macOS, Linux and Windows Server. Read-only connectors cover AWS, Azure and Google Cloud, and other sources such as identity and code.

Finds what nobody approved

AI command-line tools and agents, provider keys, tool servers, coding-assistant plug-ins, AI libraries, apps, browser extensions, traffic to AI services, AI-generated repositories and the identities agents use.

Shows what each one can reach

Files, credentials, cloud accounts and data stores, followed from the endpoint into the cloud and back, so a stray tool is shown as the path it really is.

A score you can argue with

Every rating is likelihood times impact from recorded facts. Each fact shows its proof and what would remove it, and a missing fact counts as nothing.

A decision for every item

An owner, an approved or not-allowed decision with conditions, and a review date that returns the item to its owner when it passes.

Evidence you can hand over

Signed reports with control mapping for NIST AI RMF, ISO/IEC 42001, the OWASP Top 10 for LLM applications and the EU AI Act. This is mapping and evidence, not a certification.

AI-SPM in plain words

Cloud posture tools start from cloud accounts and their configuration. AI does not wait for a cloud account: it shows up as an assistant on a laptop, a key in a shell profile or a tool server on a developer machine. AI-SPM covers that layer and connects it to what you already run. It adds to your other security tools and does not replace them.

FAQ

Common questions about AI-SPM

What is AI security posture management (AI-SPM)?

It is the practice, and the tooling, of continuously finding the AI an organisation runs, working out how exposed it is, and governing it against policy, so AI risk is measured and not guessed.

How is it different from a cloud security posture tool?

A cloud posture tool starts from cloud accounts and their configuration. AI-SPM also has to find AI that never touches a cloud account, such as assistants on laptops, keys in shell profiles and tool servers on a developer machine. Sargha starts from the endpoint and the cloud and connects them. The two work together.

Does Sargha read prompts or file contents?

No. The endpoint sensor never reads prompts, responses, file contents or keystrokes, and key values are masked where they are collected. The optional managed browser extension sends only the text a person is about to submit, over TLS, to your own collector, where it is evaluated in memory and not stored.

Is Sargha SOC 2 certified, and does it certify my organisation?

SOC 2, ISO/IEC 27001 and CSA STAR are in progress; none is attested or certified today. Sargha produces control mapping and signed evidence. It does not certify any organisation.

What does it not do?

Coverage is never 100%, and the blind spots are listed. Discovery is read-only and observe-and-warn by default; response is an opt-in, audited mode. Planned capabilities are on the roadmap and are not available today.

What you need

A free assessment covers up to 25 endpoints, read-only, and returns a risk report and an AI Bill of Materials in about a week.
Coverage is measured and the blind spots are named. It is never claimed as 100%.

For your technical team

Sensor, connector and data-handling detail is in the technical pack, shared with your team under NDA.

Request the demo and technical pack

Get started

See the AI on your fleet in about a week.

Run the sensor across your estate and receive an AI risk report and an AI Bill of Materials. The assessment is free and carries no commitment.

Request a free assessment → Book a 30-minute call