AI subscriptions are becoming real business infrastructure. We use them to write, research, build, analyse and serve customers. That makes one question non-negotiable: when usage is charged against an account, can the account holder see where it went?
This summer I had a problem with my Claude Max 20x account. The usage dashboard rose substantially during periods when I had not deliberately started work. I documented screenshots and timestamps, contacted support, and asked for an itemised explanation of the activity being counted.
The point is simpler: I could see the balance change, but I could not see an account-level record explaining the change.
What I recorded
I kept dated screenshots and a short activity log. This anonymised extract shows the kind of evidence a customer can preserve without publishing prompts, account identifiers or private business information.
| Time | Account activity | Usage view |
|---|---|---|
| 10:27 | The dashboard was checked before beginning deliberate Claude activity for that period. Other authenticated sessions, devices and connected tools could still have existed. | All-model weekly allowance showed 45% used. |
| 16:22 | Usage dashboard checked again; no deliberate work was recorded in the intervening period. | All-model weekly allowance showed 55% used. |
This alone does not establish the reason for the change. It does establish why an itemised record was needed to investigate it properly.
Why this matters beyond one account
Modern AI products can share a single allowance across web, desktop applications, coding tools, integrations and automated features. That can be convenient, but it also creates an attribution problem. A user may know that a limit has been reached without knowing which surface, session, model, device or automation used it.
For a casual experiment, that lack of detail is frustrating. For a paying customer running a business, it makes cost control and security investigation difficult. If an allowance unexpectedly falls, the customer needs enough evidence to answer basic questions:
- When did the consumption occur?
- Which product surface used it?
- Which model and session were involved?
- Which device or integration authenticated the activity?
- Was it a user-initiated request, a background task, or something else?
Those are not unreasonable demands. They are the equivalent of an itemised phone bill or cloud-usage ledger.
What to investigate before drawing conclusions
There are several neutral possibilities worth checking when an AI allowance changes unexpectedly. These are diagnostic possibilities, not conclusions about my account or any provider:
- another authenticated device, browser or session;
- a coding tool, agentic workflow or other product surface drawing from the same allowance;
- connected integrations, extensions or scheduled tasks;
- background or automated activity;
- delayed or changed dashboard reporting; or
- unauthorised account access.
Anthropic’s own guidance explains how to review active sessions and, where appropriate, log out of all sessions. Its documentation also states that some Claude plan usage is combined across Claude conversations and Claude Code terminal usage. That is why a structured, time-bound record matters.
The practical lesson
If you use AI tools for work, treat them as part of your operating system rather than as an isolated chat window:
- Take dated screenshots of usage dashboards before and after significant work.
- Keep a short activity log when a problem begins.
- Review connected apps, authorisation tokens, devices and scheduled tasks.
- Change passwords and revoke access if there is any genuine concern about account access.
- Ask support for an itemised, time-bound usage record rather than a general explanation.
The strongest outcome is not an argument about a single percentage meter. It is better observability: a customer should be able to reconcile usage with real activity.
The standard worth expecting
For every material usage event, a customer should be able to see a timestamp, product surface, model or service class, session/device or authorised integration identifier, whether activity was direct or automated, and a plain-language explanation of how it contributed to a limit.
Sensitive information can still be protected. The goal is not to expose private prompts or provider internals. It is to give customers enough information to manage costs, investigate anomalies and use AI with confidence.
As AI becomes business infrastructure, usage transparency should be treated as a basic operational control—not an optional dashboard feature.
Need clarity across your AI tools?
When a business uses multiple AI products, permissions, connectors and automated workflows, it needs a clear map of what is connected and what is running. AI Fusion Automations can help you review that operational picture in an AI Systems Snapshot.
Explore the AI Systems SnapshotRelated reading
- Restored Access Is Not the Same as Usage Attribution
- What an AI operating system can look like for a growing business
- Free business tools that can save service companies time
Official resources
- Anthropic: Manage usage credits for paid Claude plans
- Anthropic: Managing your active sessions
- Anthropic: How to log out of all active sessions
This is a first-person account of my own experience. It does not allege that Anthropic has confirmed a product defect, an account compromise, malware, or a cause connected to a particular application.