Action Risk Score
Action Risk Score for AI Agent Actions
An Action Risk Score is Asyrx's assessment of how concerning a proposed AI-agent action is given the context surrounding that specific action.
The score helps distinguish routine, expected activity from unusual, sensitive, destructive, high-value or otherwise consequential actions.
Asyrx evaluates the proposed action before execution and returns structured decision-support intelligence.
The question Action Risk answers
How concerning is this specific proposed action given what Asyrx currently knows?
Asyrx reads the structured context supplied with the proposed action. The signals below may contribute to an Action Risk assessment depending on what the calling system provides and how the organization is configured.
Financial value
The monetary value of the proposed action and whether it sits inside the expected range.
Entities affected
How many records, customers, accounts, orders or resources the action would touch.
Action type
What the agent is proposing to do — read, update, export, delete, refund, deploy, grant.
Destination
Where data, funds or authority would move, and whether the destination is internal or external.
Known vs novel destination
Whether the destination or recipient is established in the agent's recent history.
Data classification
Whether confidential, restricted or regulated data is in scope.
Production environment
Whether the action targets live production systems.
Reversibility
Whether the action can be undone, and how difficult recovery would be.
Privilege change
Whether the action expands permissions, roles, ownership or credentials.
Action velocity
Whether the agent's recent action rate is well above its normal pace.
Historical agent behavior
Whether the action is consistent with what this agent normally proposes.
Policy conditions
Whether a configured organization policy condition or threshold is implicated.
Unusual action scope
Whether the scope of the action is atypical for the agent, entity or system.
Sensitive-data involvement
Whether sensitive data is accessed, modified, exported or shared.
Asyrx publishes what the assessment considers, not how it is weighted. Signal weighting, behavioral baselines and anomaly thresholds are part of the Asyrx implementation.
Visualising the assessment
Proposed action, context signals, Action Risk
INPUT
Proposed action
An agent proposes an action. The enterprise or platform sends structured metadata describing it to Asyrx before anything executes.
CONTEXT
Context signals
Scope, value, destination, data classification, environment, reversibility, privilege, velocity, behavioral history and policy conditions.
OUTPUT
Action Risk
A score, a level, a confidence value, risk factors, reason codes and recommended controls returned to the caller.
Scale
Action Risk is reported on a 0–100 scale
These are the risk bands implemented by the current Asyrx assessment engine, and they are the same bands used across the API, the application and the assessment record.
| Score range | Level | Meaning |
|---|---|---|
| 0 – 39 | Low | The available context produces a relatively low risk assessment. |
| 40 – 64 | Moderate | Context contains signals worth noting before execution. |
| 65 – 84 | High | Multiple or severe signals are present in the action context. |
| 85 – 100 | Critical | The context indicates a highly concerning proposed action. |
Worked examples
Routine activity next to a consequential export
Every score on this page is produced by the same deterministic Asyrx assessment engine that powers the API, from the same demo action inputs. Nothing here is a marketing number.
Example · routine
Update three expected CRM records
An agent updates three account status fields as part of its normal daily workflow.
- Normal action type for this agent
- Expected record volume
- Known target system and resource
- No sensitive-data export
- No privilege change
- Limited scope
Action Risk
10 / 100
PRODUCTION_ENVIRONMENTInterpretation: Low Action Risk. Low does not mean guaranteed safe — it means the available context produces a relatively low Asyrx risk assessment.
Recommended control: No Additional Control Suggested. Asyrx recommends; it does not approve, deny, block or execute.
Example · consequential
Export 84,000 confidential CRM records
The same agent proposes exporting the full contact database to a new external analytics destination.
- High entity count
- Sensitive-data export
- Destination novelty
- External destination
- Production context
Action Risk
60 / 100
NOVEL_DESTINATIONHIGH_ENTITY_COUNTSENSITIVE_DATA_EXPORTPRODUCTION_ENVIRONMENTInterpretation: Action Risk 60 (moderate) with a Blast Radius of 97 (critical) — the action's context is elevated, and its potential consequence is larger still.
Recommended control: Secondary Approval Recommended. Asyrx recommends; it does not approve, deny, block or execute.
Severity changes the score. A proposed deletion of 25,000 production records from the same agent returns Action Risk 80 (high) with a Blast Radius of 96 (critical), because destructive and low-reversibility signals are present alongside the volume.
Risk is different from permission
Action Risk is not authorization
Identity asks
Who is acting?
Authorization asks
Can the identity perform the operation?
Action Risk asks
How concerning is this specific proposed action?
An agent may have legitimate permission and still propose an action whose context warrants additional scrutiny. Asyrx does not replace IAM, authorization or policy enforcement — it adds an action-level risk perspective those systems do not provide.
Two separate values
Action Risk and Confidence are not the same thing
Action Risk
How concerning the proposed action appears given its context.
Confidence
How confident Asyrx is in the assessment given the context available. It is not a probability of fraud, misuse or loss.
High Action Risk · higher confidence
The action context is concerning and Asyrx had substantial supporting context — behavioral history, destination history and complete action metadata.
High Action Risk · lower confidence
The action context still looks concerning, but important context was missing. Asyrx reports the gap instead of implying precision it does not have.
Outputs
What an Action Risk assessment may include
Exact fields depend on the endpoint, the action context supplied and the organization's configuration. The API reference is authoritative.
Score
Action Risk on a 0–100 scale.
Level
Low, moderate, high or critical, derived from the same scale.
Confidence
How confident Asyrx is in the assessment given the context supplied.
Reason codes
Stable, machine-readable identifiers for each contributing signal.
Risk factors
Human-readable labels, categories and severities behind those codes.
Behavioral context
Comparison against the agent's recent activity where history exists.
Policy context
Which configured policy conditions were implicated, where configured.
Recommended controls
Advisory safeguards such as verification, human review or secondary approval.
Where Action Risk sits
From proposed action to enterprise decision
STEP 1
High-impact action
The consequential action the agent proposes.
STEP 2
Action Risk Score
How concerning is this specific action in context?
STEP 3
Blast Radius Score
How large could the potential consequence be?
STEP 4
Risk factors & reason codes
Why Asyrx considers the action concerning.
STEP 5
Recommended controls
Which additional safeguards may be appropriate.
STEP 6
Customer decision
The enterprise retains authorization and execution control.
Action Risk and Blast Radius can diverge. A privilege change affecting a single account returns Action Risk 28 (low) with a Blast Radius of 85 (critical) — modest scope, large potential consequence.
FAQ
Action Risk questions
Explore Asyrx risk intelligence
Related Asyrx concepts
Action Risk, Blast Radius and high-impact action categories are three parts of the same assessment.