Overview

SurePath AI seamlessly integrates with your preferred cloud file storage providers to deliver detailed User Activity logs and Administrative audit trail events. These logs are designed for easy integration with your SIEM or other log management and ingestion tools, enabling centralized monitoring and analysis.

The logs provide a comprehensive record of user interactions across various services, including policy enforcement outcomes, sensitive data handling events, user intents, service responses, and risk assessments. This rich data supports auditing, troubleshooting, and analysis, empowering your tools with insights for alerting, reporting, and incident response.

To configure where these logs are delivered, see Configuring telemetry destinations.

File information

Retention

SurePath AI does not manage file deletion from your cloud storage. Ensure your organization implements cleanup procedures according to its data retention policies.

Output frequency

User activity and related information are uploaded every 15 minutes into new files.

File naming convention and directory structure

Log files are organized in a hierarchical directory structure based on UTC timestamps and use clear naming to indicate the time range of the data they contain.

Directory structure:
User Events: surepath-ai/user-events/v2/YYYY/MM/DD/HH/

Audit Events: surepath-ai/audit-events/v1/YYYY/MM/DD/HH/

Files are partitioned by UTC hour, making it easy to locate events from specific time periods. If your organization has configured a custom bucket key prefix, it will be prepended to this path.

File naming pattern: <start-timestamp>-<end-timestamp>-part-<number>.ndjson.gz

Each file name includes:

Start timestamp: The earliest event timestamp in the file
End timestamp: The latest event timestamp in the file
Part number: A zero-padded six-digit chunk identifier (e.g., part-000001)

Example path: surepath-ai/user-events/v2/2025/10/09/15/2025-10-09T15-07-57-875Z-2025-10-09T15-08-45-123Z-part-000001.ndjson.gz

File format and compression

Files are delivered in NDJSON (newline-delimited JSON) format, where each line represents a single JSON event object. This format enables efficient streaming and processing of large datasets.

All files are compressed using gzip compression to reduce storage costs and transfer times. Files are automatically chunked into parts of up to 10,000 lines each to maintain optimal file sizes for processing.

Log format

The log format of User Events and Audit events consist of NDJSON, adhering to standard JSON syntax and conventions. Each line in the file represents a complete JSON object containing detailed logs of user activities and system interactions.

Example of a single event entry:

{  "event": {    "id": "evt-123",    "category": "user",    "type": "intercept",    "action": "allow",    "schema_version": "v2.0.1",    "timestamp": "2025-10-09T15:07:57.875Z",    "trace_id": "abc123"  },  "destination": {    "name": "ChatGPT"  },  "actor": {    "name": "Jane Doe",    "email": "[email protected]",    "type": "user"  },  "http": {    "url": "https://api.example.com/intercept",    "user_agent": "Mozilla/5.0"  },  "network": {    "remote_ip": "203.0.113.10",    "remote_port": "443",    "internal_ip": "10.0.0.5",    "x_forwarded_for": "203.0.113.10"  },  "policy": {    "decision": "allow",    "violations": {      "access": false,      "pii": false,      "intent": false,      "confidential_data": false,      "prompt_injection": false,      "toxicity": false,      "code": false,      "bias": false,      "risk": false    }  },  "conversation": {    "id": "conv-1"  },  "messages": {    "input": [      {        "role": "user",        "content": "Create a social media post about our new feature"      }    ],    "output": [      {        "role": "assistant",        "content": "Here's a draft post..."      }    ]  },  "intent": {    "domain": "Marketing",    "action": "create social media post"  },  "risk": {    "overall": "medium",    "destination": "low",    "input": {      "overall": "medium",      "intent": "low",      "data_sensitivity": "internal",      "data_exposure_impact": "medium",      "harmful_content": "low",      "prompt_injection": "low"    },    "output": {      "overall": "low",      "harmful_content": "low",      "bias": "low"    }  },  "detections": [    {      "type": "pii",      "name": "person_name",      "action": "mask"    }  ],  "gen_ai": {    "model_name": "gpt-4o",    "model_id": "gpt-4o-2024-06",    "assistant_name": "Analyst",    "token_count": {      "input": 125,      "output": 98    }  },  "timing": {    "downstream_start": "2025-10-09T15:07:57.900Z",    "downstream_end": "2025-10-09T15:07:58.100Z"  }}

User Events - JSON field descriptions

The following tables describe the structure and fields available in the telemetry schema. Not all fields are present in every event.

Event metadata

Field	Type	Description
`event`	object	Event identity, classification, and timing
`event.id`	string	Unique event identifier
`event.category`	string	Event category (e.g., user, audit, network)
`event.type`	string	Event type (e.g., intercept, access, internal)
`event.action`	string	Event action (e.g., allow, block, redirect, redact, login, error)
`event.schema_version`	string	Full schema version string (e.g., v2.0.1)
`event.schema_url`	string	URL where the schema can be referenced
`event.timestamp`	string	ISO 8601 UTC timestamp
`event.trace_id`	string (uuid)	Correlation identifier for the event that can be provided to SurePath AI support for investigation

Destination

Field	Type	Description
`destination`	object	Information about the destination service
`destination.name`	string	Destination service name from the Public Service Catalog

Actor

Field	Type	Description
`actor`	object	Actor (user or app) associated with the event
`actor.name`	string	Actor name
`actor.email`	string	Actor email address
`actor.type`	string	Actor type (e.g., user, app)

HTTP and network

Field	Type	Description
`http`	object	HTTP request information
`http.url`	string	Request URL
`http.user_agent`	string	User agent string
`network`	object	Network connection information
`network.remote_ip`	string	Remote address IP
`network.remote_port`	string	Remote address port
`network.internal_ip`	string	Internal IP address
`network.x_forwarded_for`	string	X-Forwarded-For header value indicating IP addresses from which a client has been forwarded through proxies

Policy decision

Field	Type	Description
`policy`	object	Policy decision and violation information
`policy.decision`	string	Policy decision (e.g., allow, block, redirect, redact)
`policy.violations`	object	Violation flags
`policy.violations.access`	boolean	Access violation detected
`policy.violations.pii`	boolean	PII violation detected
`policy.violations.intent`	boolean	Intent violation detected
`policy.violations.confidential_data`	boolean	Confidential data violation detected
`policy.violations.prompt_injection`	boolean	Prompt injection violation detected
`policy.violations.toxicity`	boolean	Toxicity violation detected
`policy.violations.code`	boolean	Code-sharing violation detected
`policy.violations.bias`	boolean	Bias violation detected
`policy.violations.risk`	boolean	Risk policy violation detected

Conversation and messages

Field	Type	Description
`conversation`	object	Conversation context
`conversation.id`	string	Conversation identifier
`messages`	object	Input/output message content captured for the event
`messages.input[]`	array	Array of input message items
`messages.input[].role`	string	Message role (e.g., user, assistant, system)
`messages.input[].content`	string	Message text content. Detected PII data may be replaced with tags (e.g., [PERSON], [US_SSN])
`messages.output[]`	array	Array of output message items with the same shape as input

Intent classification

Field	Type	Description
`intent`	object	High-level domain and action classification
`intent.domain`	string	Domain classification (e.g., Human Resources, IT, Finance, Marketing)
`intent.action`	string	Action classification (e.g., generate, execute, analyze, create)

Risk assessment

Field	Type	Description
`risk`	object	Risk assessment across overall, input, output, and destination
`risk.overall`	string	Overall risk rating (low, medium, high)
`risk.destination`	string	Destination-specific risk rating (low, medium, high)
`risk.input`	object	Input risk assessment
`risk.input.overall`	string	Overall input risk rating (low, medium, high)
`risk.input.intent`	string	Input intent risk (low, medium, high)
`risk.input.data_sensitivity`	string	Data sensitivity level (unknown, public, internal, confidential, critical)
`risk.input.data_exposure_impact`	string	Potential impact if data were exposed (low, medium, high)
`risk.input.harmful_content`	string	Harmful content risk (low, medium, high)
`risk.input.prompt_injection`	string	Prompt injection risk (low, medium, high)
`risk.output`	object	Output risk assessment
`risk.output.overall`	string	Overall output risk rating (low, medium, high)
`risk.output.harmful_content`	string	Harmful content risk (low, medium, high)
`risk.output.bias`	string	Bias risk (low, medium, high)

Detections

Field	Type	Description
`detections[]`	array	Detection results associated with the event
`detections[].type`	string	Detection type (e.g., pii, code, intent)
`detections[].name`	string	Detection name (e.g., person_name, code_block, custom intent)
`detections[].action`	string	Detection action taken (e.g., mask, delete, block)

Generative AI metadata

Field	Type	Description
`gen_ai`	object	Generative AI metadata for the event
`gen_ai.model_name`	string	Model name
`gen_ai.model_id`	string	Model identifier
`gen_ai.assistant_name`	string	Assistant name
`gen_ai.token_count`	object	Token usage information
`gen_ai.token_count.input`	integer	Input token count
`gen_ai.token_count.output`	integer	Output token count

Timing and errors

Field	Type	Description
`timing`	object	The start and end timestamps for request processing
`timing.downstream_start`	string	Downstream processing start timestamp (ISO 8601 UTC)
`timing.downstream_end`	string	Downstream processing end timestamp (ISO 8601 UTC)
`error`	object	Error context for the event
`error.type`	string	Error type (e.g., service-error, subscription-expired, maintenance)

Audit Events - JSON field descriptions

The following tables describe the structure and fields available in the audit trail schema. Audit trail events record administrative actions or system events performed within the SurePath AI platform. Not all fields are present in every event.

Event metadata

Field	Type	Description
`_id`	string	Unique event identifier
`eventAction`	string	Administrative action performed (e.g., upload, read)
`eventCategory`	string	Event category (e.g., audit)
`eventDateTime`	string	ISO 8601 UTC timestamp of when the action occurred
`eventType`	string	Type of administrative action (e.g., audit-event-sync, Partial Conversation, Full Conversation)
`message`	string	Human-readable description of the administrative action
`orgId`	string	Organization identifier
`version`	string	Schema version (e.g., v1)

Resource

Field	Type	Description
`resourceId`	string	Identifier of the resource the administrator acted upon
`resourceName`	string	Name of the resource the administrator acted upon

Actor

Field	Type	Description
`actor`	object	Administrator who performed the action
`actor.assumed`	boolean	Whether the administrator's identity was assumed (impersonated)
`actor.email`	string	Administrator's email address
`actor.name`	string	Administrator's display name
`actor.orgId`	string	Administrator's organization identifier
`actor.userId`	string	Administrator's user identifier

Additional properties

Field	Type	Description
`additionalProperties`	object	Supplemental context for the administrative action
`additionalProperties.resourceOwner`	object	End user who owns the resource being accessed by the administrator
`additionalProperties.resourceOwner.id`	string	Resource owner's user identifier
`additionalProperties.resourceOwner.name`	string	Resource owner's display name
`additionalProperties.resourceOwner.email`	string	Resource owner's email address

Migrating to a newer version of the schema

To migrate to a newer version of the schema a new Telemetry Destination must be added. Previous Telemetry Destinations with older versions of the schema will continue to function while customers integrate the new schema into their telemetry destination.

For AWS S3 Buckets customers can use the same connector and bucket name in the new Telemetry Destination as newer schemas contain different prefixes allowing two versions of schema to be written to the same location. Other telemetry connectors' functionality may differ.

When the newer schema integration is validated, the previous Telemetry Destination can be disabled and then eventually deleted.

Legacy V1 schema (Deprecated)

The information below describes the legacy V1 schema format. New telemetry destinations automatically use the latest V2 schema described above. Existing destinations using V1 continue to operate with their configured version.

V1 file information

Directory structure: user-events/

V1 uses a flat directory structure without hierarchical date-based partitioning. If your organization has configured a custom bucket key prefix, it will be prepended to this path.

File naming pattern: <start-timestamp>-<end-timestamp>.json

Each file name includes the earliest and latest event timestamps in the file using ISO 8601 UTC format.

Example path: user-events/2024-12-10T16-48-29-028Z-2024-12-10T16-48-45-354Z.json

V1 file format

Files are delivered as a single JSON array containing multiple event objects. V1 files are not compressed and do not use chunking. Each file contains a complete JSON array with all events from the 15-minute export window.

V1 format example

[  {    "traceId": "d1a2b3c4",    "serviceName": "ChatGPT",    "messages": {      "request": [{ "role": "user", "content": "Summarize quarterly results" }],      "response": [{ "role": "assistant", "content": "Here is a summary..." }]    },    "intent": {      "request": [{ "action": "Create social media post", "role": "Marketing" }]    },    "sensitiveData": { "request": { "action": "mask" } },    "policyDecision": {       "action": "allow",       "code": "P200",       "violations": { "sensitiveData": false, "access": false }     },    "firewallDecision": { "target": { "name": "Engineering (gpt-4o)" } },    "user": { "name": "Jane Doe", "email": "[email protected]" },    "userAgent": "Mozilla/5.0",    "clientIp": {       "remoteIp": "203.0.113.10",       "remotePort": "443",       "xForwardedFor": "203.0.113.10"     },    "startTime": 1733246296533,    "endTime": 1733246335420,    "duration": "00:00:39.887",    "requestUrl": "https://api.example.com/analyze"  }]

V1 JSON fields

Field	Type	Description
`traceId`	string (uuid)	Correlation identifier for the event used to trace requests across systems that can be provided to SurePath AI support for investigation
`serviceName`	string	Name of the destination service name from the Public Service Catalog
`messages`	object	Input/output message content for the event
`messages.request[]`	array	Array of message items
`messages.request[].role`	string	Message role (e.g., user, assistant, system)
`messages.request[].content`	string	Message text content
`messages.response[]`	array	Array of message items with the same shape as request
`intent`	object	Detected user intent information
`intent.request[]`	array	Array of intent entries
`intent.request[].action`	string	Intent action label
`intent.request[].role`	string	Role associated with the intent
`sensitiveData`	object	Sensitive data handling outcome for the request
`sensitiveData.request`	object	Sensitive data request details
`sensitiveData.request.action`	string	Sensitive data action taken (e.g., delete, mask, synthesize, tag)
`policyDecision`	object	Decision and metadata from the policy engine
`policyDecision.action`	string	Policy decision action (e.g., allow)
`policyDecision.code`	string	Policy decision code for categorization that can be provided to SurePath AI support for investigation
`policyDecision.violations`	object	Violation flags
`policyDecision.violations.sensitiveData`	boolean	Indicates a sensitive data violation
`policyDecision.violations.access`	boolean	Indicates an access violation
`firewallDecision`	object	Model routing decision details
`firewallDecision.target`	object	Target model information
`firewallDecision.target.name`	string	Selected target/model name as defined by admins in the SurePath AI console
`user`	object	Actor identifiers
`user.name`	string	End-user display name
`user.email`	string	End-user email address
`userAgent`	string	Client user agent string
`clientIp`	object	Client network identifiers
`clientIp.remoteIp`	string	Remote address IP
`clientIp.remotePort`	string	Remote address port
`clientIp.xForwardedFor`	string	X-Forwarded-For header value
`startTime`	integer	Millisecond epoch when processing started
`endTime`	integer	Millisecond epoch when processing ended
`duration`	string	Duration of processing (formatted as HH:MM:SS.mmm)
`requestUrl`	string	Full request URL

Exporting User Activity logs

Sensitive Data policy

Reading User Activity events

Configuring telemetry destinations

Telemetry schema and format

Overview

File information

Retention

Output frequency

File naming convention and directory structure

File format and compression

Log format

User Events - JSON field descriptions

Event metadata

Destination

Actor

HTTP and network

Policy decision

Conversation and messages

Intent classification

Risk assessment

Detections

Generative AI metadata

Timing and errors

Audit Events - JSON field descriptions

Event metadata

Resource

Actor

Additional properties

Migrating to a newer version of the schema

Legacy V1 schema (Deprecated)

V1 file information

V1 file format

V1 format example

V1 JSON fields