PAM API

This reference covers the Privileged Access Management (PAM) REST surface: the admin control plane under /api/v1/pam, the device actuation endpoint, and the agent-side ingest endpoint. For the administrative workflow these endpoints back, see Privileged Access Management; for the trust model, see the PAM Security Model.

All paths are relative to the API base URL and version prefix (/api/v1). See the API Overview for authentication, base URL, and common conventions.

Authentication and authorization

The /pam/* control-plane endpoints require a JWT bearer token (or API key) and run under organization, partner, or system scope. Authorization maps to the device permission set:

Operation class	Permission	Extra gate
Read (list, active, rules)	devices:read	–
Rule create / update / delete	devices:write	MFA
Approve / deny / revoke	devices:execute	MFA

Tenant isolation is enforced by row-level security on every query. Site-restricted technicians are additionally narrowed to their allowed sites: requests outside those sites are filtered from reads and rejected on writes with 403.

Note

Every mutating call records an entry in both the platform audit log and the dedicated elevation audit ledger. Approvals, denials, and revocations are attributed to the calling user by id.

---

Elevation requests

Method	Path	Description
GET	/pam/elevation-requests	List and filter elevation requests (Requests and Audit tabs).
GET	/pam/active	List currently-active (unexpired, approved) elevations.
POST	/pam/elevation-requests/:id/respond	Approve or deny a pending request.
POST	/pam/elevation-requests/:id/revoke	Revoke an active elevation before it expires.

List elevation requests

GET /pam/elevation-requests

Query parameters (all optional):

Param	Type	Notes
status	enum	pending, approved, auto_approved, denied, expired, revoked, actuating
flowType	enum	uac_intercept, tech_jit_admin, ai_tool_action
deviceId	uuid	Filter to one device
siteId	uuid	Filter to one site (must be within the caller’s site scope)
from / to	ISO-8601	Bound by request time
page / limit	int	limit max 100, default 50

Response:

{
  "success": true,
  "requests": [
    {
      "id": "…",
      "orgId": "…",
      "deviceId": "…",
      "deviceHostname": "WIN-ACCT-04",
      "siteName": "HQ",
      "flowType": "uac_intercept",
      "status": "pending",
      "subjectUsername": "acme\\sarah",
      "targetExecutablePath": "C:\\Users\\sarah\\Downloads\\FirefoxSetup.exe",
      "targetExecutableSigner": "Mozilla Corporation",
      "requestedAt": "2026-06-13T14:02:11Z",
      "approvedByName": null,
      "deniedByName": null,
      "revokedByName": null,
      "matchedPolicyName": null,
      "pamRuleId": null,
      "pamRuleName": null,
      "decisionSource": null
    }
  ],
  "pagination": { "page": 1, "limit": 50, "total": 1 }
}

decisionSource is derived: software_policy, pam_rule, human, or null (still pending). AI tool-action rows additionally carry toolName, riskTier, and actionDigest.

List active elevations

GET /pam/active

Returns up to 500 elevations whose status is approved, auto_approved, or actuating and whose expiresAt is in the future. No query parameters. Used to render the live “active windows” view with expiry countdowns.

{ "success": true, "active": [ { "id": "…", "expiresAt": "2026-06-13T14:17:00Z", "…": "…" } ] }

Respond (approve / deny)

POST /pam/elevation-requests/:id/respond — requires devices:execute + MFA.

Request body:

{
  "decision": "approve",
  "reason": "Approved per change CHG-1042",
  "durationMinutes": 30
}

Field	Type	Notes
decision	enum	approve or deny (required)
reason	string	Optional on approve, recommended on deny; max 2000 chars
durationMinutes	int	Approval window, 1–1440 (24h max). Defaults to 15

Only a pending request can be decided. The transition is an atomic compare-and-swap on pending, so concurrent decisions are safe.

Status	Meaning
200	{ "success": true, "id": "…", "status": "approved" | "denied" }
400	Invalid id or body
403	Site access denied, or MFA not satisfied
404	Request not found (or outside the caller’s org)
409	Request is no longer pending (lost the race), or a linked AI tool execution already timed out

Note

For an ai_tool_action request, an approval is mirrored onto the linked AI tool execution inside the same transaction. If that execution already timed out, the whole decision is rolled back and the call returns 409 rather than recording a grant that can never take effect.

Revoke

POST /pam/elevation-requests/:id/revoke — requires devices:execute + MFA.

Ends an active elevation early.

{ "reason": "Maintenance window cancelled" }

reason is required (1–2000 chars). Only an active elevation (approved, auto_approved, or actuating) can be revoked; a stale or already-finished row returns 409.

{ "success": true, "id": "…", "status": "revoked" }

---

Rules

Method	Path	Description
GET	/pam/rules	List rules in priority order.
POST	/pam/rules	Create a rule.
POST	/pam/rules/preview	Dry-run draft criteria against request history.
PATCH	/pam/rules/:id	Update a rule.
DELETE	/pam/rules/:id	Delete a rule.

Create / update a rule

POST /pam/rules (create) and PATCH /pam/rules/:id (partial update) — require devices:write + MFA.

{
  "name": "Auto-approve signed Mozilla installers",
  "verdict": "auto_approve",
  "priority": 50,
  "enabled": true,
  "matchSigner": "Mozilla Corporation",
  "approvalDurationMinutes": 30,
  "siteId": null
}

Field	Type	Notes
name	string	Required, 1–255 chars
verdict	enum	auto_approve, auto_deny, require_approval, ignore
priority	int	Lower runs first; default 100
enabled	bool	Default true
siteId	uuid | null	Narrow to one site; null = org-wide
matchSigner	string	Code-signer (executable rule). Mutually exclusive with matchSignerGroupId
matchSignerGroupId	uuid | null	Reference a reusable signer group instead of a single matchSigner (executable rule)
matchHash	string	SHA-256 hex digest (executable rule)
matchPathGlob	string	File-path glob (executable rule)
matchParentImage	string	Parent process (executable rule)
matchCommandLine	string	Case-insensitive substring of the launched process command line (executable rule)
matchUser	string	Narrow by requesting user
matchAdGroup	string	Narrow by AD group
matchToolName	string	AI tool name (tool-action rule)
matchRiskTier	int	Risk tier 0–4, matched by exact equality (tool-action rule). See Risk-tier drift
matchNegate	string[] | null	Criterion keys the engine inverts (“does not match”). Valid keys: signer, signerGroup, hash, pathGlob, parentImage, commandLine, user, adGroup, toolName, riskTier
timeWindow	object	{ start, end, days?, timezone? }, HH:MM 24h, days 0–6 (Sun–Sat)
approvalDurationMinutes	int	1–1440; null falls back to the org default

Rule-shape validation (enforced on create and on the merged result of an update):

• A rule must carry at least one identifying criterion (signer / signer group / hash / path / parent / command line / user / group / tool / tier). A rule scoped only by time window – or nothing – is rejected with 400.
• A rule is either executable-shaped (signer / signer group / hash / path / parent / command line) or tool-action-shaped (tool name / risk tier). Mixing the two is rejected, because no single observation carries both.
• matchSigner and matchSignerGroupId are mutually exclusive – a rule references a single signer or a signer group, not both.
• A negated criterion never turns missing data into a match: if the observation didn’t report the value a negated criterion checks, the rule does not match.
• verdict: "ignore" is not valid for tool-action rules – a tool action must be decided.
• matchRiskTier must be a tier the rule’s tool selector can actually resolve to. A rule pinned to a tier no tool produces can never match, so create and update reject it with 400 and a body carrying code: "pam_rule_risk_tier_unreachable" plus a validTiers array of the tiers that selector currently resolves to. A rule naming a tool the API doesn’t recognise (an extension tool, for example) is not rejected – the check fails open rather than blocking a legitimate write.

Create returns 201 with the full rule; update returns 200.

Risk-tier drift

matchRiskTier is compared with exact equality, and a tool’s risk tier is decided by the API build, not by the rule. So when a release re-classifies a tool – as one did when three read-only execute_command command types moved from Tier 3 to Tier 2 – a stored rule written against the old tier silently stops covering those calls.

The failure direction is safe: an unmatched tool action falls through to the org’s default (pending / require approval) rather than auto-approving. But it is invisible, so the API surfaces it three ways:

• On write – create and update reject a tier the selector can never resolve to (above).
• At boot – the API scans enabled, tier-pinned rules on startup and logs a [pam-rule-tier-drift] warning naming every rule that can no longer match, alongside a Sentry event tagged pam_rule_risk_tier_unreachable. A deploy is the only moment a stored rule can go stale, so this runs once per start rather than on a schedule.
• In the list – every row from GET /pam/rules carries matchRiskTierStale (bool) and matchRiskTierValidTiers (number[] | null), which the console renders as a warning badge on the rule.

A rule that merely narrowed is deliberately not flagged. A rule matching execute_command at Tier 3 still covers file_read and kill_process after the re-classification – it lost some coverage but is not dead, and many tools legitimately span several tiers. Only a tier the selector can never produce is reported.

Matching is not widened to “this tier or higher” on purpose: that would move the auto-approval boundary of every existing rule, which is a policy change rather than a bug fix.

Preview a rule

POST /pam/rules/preview — requires devices:write.

Dry-runs draft criteria against historical requests so you can see what a rule would have matched before saving it. Accepts the same criteria fields as a rule, plus windowDays (1–90, default 30) and an optional flowType.

{
  "success": true,
  "totalMatched": 12,
  "totalScanned": 480,
  "windowDays": 30,
  "truncated": false,
  "statusBreakdown": { "pending": 2, "approved": 7, "auto_approved": 3, "denied": 0, "expired": 0, "revoked": 0, "actuating": 0 },
  "sample": [ { "id": "…", "requestedAt": "…", "flowType": "uac_intercept", "status": "approved" } ]
}

The preview is a pure per-rule match – it does not replay priority shadowing or the software-policy bridge. Historical rows do not store AD groups, so any draft containing matchAdGroup reports zero matches.

---

Signer groups

A signer group is a named, reusable set of trusted code-signer (subject) patterns that executable rules can reference via matchSignerGroupId instead of repeating the same publisher across rules.

Method	Path	Description
GET	/pam/signer-groups	List the org’s signer groups.
POST	/pam/signer-groups	Create a signer group.
PATCH	/pam/signer-groups/:id	Partial update (name, description, signers).
DELETE	/pam/signer-groups/:id	Delete a group; 409 if it is still referenced by a rule.

Reads require devices:read; writes require devices:write + MFA.

{
  "name": "Trusted browser vendors",
  "description": "Publishers whose installers we auto-approve",
  "signers": ["Mozilla Corporation", "Google LLC"]
}

Field	Type	Notes
name	string	Required, 1–255 chars
description	string | null	Optional, max 2000 chars
signers	string[]	Up to 500 trimmed entries, each max 255 chars; de-duplicated case-insensitively

GET and create/update return the full group(s):

{
  "success": true,
  "signerGroups": [
    {
      "id": "…",
      "orgId": "…",
      "name": "Trusted browser vendors",
      "description": "Publishers whose installers we auto-approve",
      "signers": ["Mozilla Corporation", "Google LLC"],
      "createdAt": "…",
      "updatedAt": "…"
    }
  ]
}

Create returns 201. DELETE returns 409 (with a message naming the referencing-rule count) when the group is still used by a rule – detach those rules first.

---

Org default verdict

The default verdict decides an elevation that matches no software policy and no PAM rule.

Method	Path	Description
GET	/pam/config	Get the org’s PAM config (devices:read).
PUT	/pam/config	Set the default unmatched verdict (devices:write + MFA).

{ "success": true, "config": { "orgId": "…", "defaultUnmatchedVerdict": "require_approval" } }

defaultUnmatchedVerdict is require_approval (the default, also applied when no config row exists) or auto_deny. PUT accepts { "defaultUnmatchedVerdict": "require_approval" | "auto_deny" }.

---

Device actuation

Method	Path	Description
POST	/devices/:id/actuate-elevation	Queue an actuate_elevation command for an approved UAC elevation.

POST /devices/:id/actuate-elevation — organization/partner/system scope, devices:execute, MFA.

Queues the agent “go” signal for a consent.exe prompt that is already waiting on the user’s screen. The command payload that reaches the agent carries only the elevation request id and a timeout – the dormant-admin credential is minted locally by the agent and never crosses the wire.

{ "elevationRequestId": "…", "timeoutMs": 8000 }

Note

The request schema also accepts optional username and password fields, but they are vestigial – the handler never reads them and never forwards them into the queued command, and they never appear in the audit log. A reader auditing the schema directly will see a password field; it is inert and exists only as a leftover of an earlier design. The credential is minted on the endpoint, not supplied over the API.

The route is single-use: it atomically transitions the request from approved to actuating in the same transaction that queues the command. A request that is not approved, or that another caller already claimed, is refused.

Status	Meaning
201	Command queued; returns the command id and elevationRequestId
400	Invalid body, or device decommissioned
403	PAM actuator disabled, site access denied, or MFA not satisfied
404	Device or elevation request not found
409	Request not approved (wrong_status) or already being actuated (race_lost)

Caution

The actuator is disabled by default in every environment and must be explicitly enabled server-side once just-in-time credentials and agent-side re-validation are deployed. While disabled, this endpoint returns 403.

---

Agent ingest

Method	Path	Description
POST	/agents/:id/elevation-requests	Agent reports an observed UAC consent prompt.

This endpoint is called by the agent, authenticated with the agent bearer token (not a user JWT), and is documented here for completeness. It records a UAC observation as an elevation_requests row with flowType: uac_intercept, runs the decisioning chain (software-policy bridge → PAM rules → org default verdict), and emits the corresponding event and audit rows. When the result is pending, Breeze additionally fans out a mobile approval request to each eligible technician and pushes it to their phones (best-effort — the 201 does not depend on push delivery; the first approver to decide on mobile wins and the siblings expire).

Request body (synthesized fields like reason are written server-side; the agent only sends discovery data):

{
  "subject_username": "acme\\sarah",
  "target_executable_path": "C:\\Users\\sarah\\Downloads\\FirefoxSetup.exe",
  "target_executable_hash": "<agent-reported file hash>",
  "target_executable_signer": "Mozilla Corporation",
  "pid": 4812,
  "parent_image": "C:\\Windows\\explorer.exe",
  "command_line": "…",
  "observed_at": "2026-06-13T14:02:11Z"
}

Note

target_executable_hash is a free-form string of up to 128 characters – the ingest schema does not validate it as hex or pin it to a 64-character SHA-256 digest. It is the value the agent reports and is stored as-is for display and rule matching. By contrast, the rule-side matchHash criterion is strictly validated as a 64-character SHA-256 hex digest, so a rule can only match on a well-formed hash.

Status	Meaning
201	{ "id": "…", "status": "pending" | "auto_approved" | "denied" }
200	{ "id": null, "status": "ignored" } – an ignore rule matched; no request row is created
413	Body exceeds 32 KB
429	Per-device rate limit exceeded (about 10 requests/second)

Decisioning fails safe: if either evaluator errors, the request is recorded as pending rather than auto-approved, regardless of the org’s configured default verdict. When no policy and no rule match (without an error), the org default verdict decides — pending under require_approval, or denied under auto_deny.

---

Configuration policy: enabling UAC capture

Whether a device captures UAC prompts at all is governed by the pam configuration-policy feature, resolved through the normal policy hierarchy (partner → org → site → group → device, closest-wins). The feature’s inline settings shape is a single flag:

{ "uacInterceptionEnabled": true }

uacInterceptionEnabled defaults to true at every level, so upgrades are behavior-preserving; admins opt a scope out by assigning a policy that turns it off. This toggle controls only whether devices capture elevation prompts – it does not govern the rule chain, request queue, or audit history, which are scoped to the organization and site and managed in the Privileged Access console. See Configuration Policies for how feature settings resolve.