Blast Radius Score
Blast Radius Score for AI Agent Actions
A Blast Radius Score is Asyrx's assessment of how large the potential consequence could be if a proposed AI-agent action is wrong, compromised, unintended or inappropriate.
Action Risk asks how concerning the action looks. Blast Radius asks how much it could cost if the action turns out to be wrong.
Both are returned together, before execution, so oversight can be directed at consequence rather than volume.
Six impact dimensions
Blast Radius is assessed across the dimensions where consequence actually lands
Each dimension is scored independently and returned with its own rationale, so a reviewer can see whether the exposure is financial, data-related, operational or a matter of privilege.
Financial impact
The monetary exposure implied by the action — value moved, refunded, committed or written off.
Entity impact
How many records, customers, accounts, orders or resources could be affected.
Data impact
The sensitivity and classification of data accessed, modified, exported or shared.
Operational impact
Whether production systems, live workflows or service availability could be disrupted.
Privilege impact
Whether permissions, roles, ownership, credentials or administrative authority could change.
Reversibility
How difficult recovery would be if the action turns out to be wrong.
Dimension breakdown
An 84,000-record confidential export, dimension by dimension
Produced by the same deterministic Asyrx assessment engine that powers the API, from the same demo action input used across the product.
Demo scenario · Salesforce bulk export
No direct monetary value attached to this action.
84,000 record(s) or entities in scope.
Data classified as confidential leaving the system of record.
Targets a production environment with live operational dependencies.
No permission or credential change detected.
Reversibility assessed as partially reversible.
The overall Blast Radius of 97 (critical) is not a single number pulled from thin air. It aggregates the six dimension scores shown here, each with its own rationale.
The corresponding Action Risk Score for the same proposed action is 60 (moderate) at 83% confidence.
That gap is the point: the action's context is elevated, while its potential consequence is larger still. Asyrx returns both values so the enterprise can weigh them.
Recommended control: Secondary Approval Recommended. Asyrx recommends; it does not approve, deny, block or execute.
Scale
Blast Radius is reported on a 0–100 scale
| Score range | Level | Meaning |
|---|---|---|
| 0 – 39 | Low | Limited potential consequence in the dimensions Asyrx can see. |
| 40 – 64 | Moderate | Meaningful consequence concentrated in one or more dimensions. |
| 65 – 84 | High | Substantial potential consequence across multiple dimensions. |
| 85 – 100 | Critical | Very large potential consequence if the action is wrong. |
Comparative examples
Low, moderate and severe consequence
Three demo actions, three very different Blast Radius outcomes — all engine-produced.
Low consequence
Routine customer refund
A small refund on a single order, inside the agent's normal value range and workflow.
Blast Radius
27 / 100
Recommended control: No Additional Control Suggested.
Moderate consequence
Large goodwill refund
A single refund well above the agent's typical value, on one customer order.
Blast Radius
79 / 100
Recommended control: Human Review Recommended.
Severe consequence
Bulk deletion of production records
A proposed deletion of 25,000 production CRM records as part of a de-duplication workflow.
Blast Radius
96 / 100
Recommended control: Secondary Approval Recommended.
Divergence
Small action, large consequence
Volume is a poor proxy for consequence, which is why Asyrx reports the two values separately.
Demo scenario · repository privilege change
One account, administrative authority
A single account is granted elevated repository access during CI onboarding. The action touches one entity.
- Action Risk
- 28
- low
- Blast Radius
- 85
- critical
Entity impact is minimal. Privilege impact and reversibility are not. A permission grant persists after the action completes and can be used to take further actions, which is why the potential consequence outruns the action's footprint.
A reviewer looking only at record counts would rank this action far below a bulk update. A reviewer looking at Blast Radius would not.
Recommended control: Secondary Approval Recommended.
Why it matters
Consequence-first oversight
Prioritise review capacity
Human attention is finite. Blast Radius directs it toward the actions where being wrong is expensive.
Explain the decision
Six dimension scores with rationale give risk, security and audit teams a defensible record of why an action was escalated.
Scale agent autonomy safely
Low-consequence actions can keep moving while consequential ones receive the controls the enterprise chooses to apply.
Where Blast Radius 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.
FAQ
Blast Radius questions
Explore Asyrx risk intelligence
Related Asyrx concepts
Action Risk, Blast Radius and high-impact action categories are three parts of the same assessment.