When AI Usage Rises While Nobody Is Working

You may not notice an AI account takeover immediately. There may be no locked office door, missing laptop, or suspicious email in your inbox. Instead, your team sees an AI usage meter rising while everyone is busy with normal work—or nobody is using the account at all.

That is the warning behind reports of Claude subscribers having their sessions stolen and their usage consumed by unauthorised users. One affected consultant reported that his token usage increased even after he disabled connected tools and stopped working. Anthropic later told affected users that infostealer malware had been used to steal login sessions, allowing attackers to access accounts and consume usage. Source: TechCrunch

For a Malaysian SME, this is not only an AI subscription problem. It is an operational risk. If your AI account is connected to customer service, accounting workflows, coding tools, documents, or internal instructions, an attacker may gain more than access to an assistant. They may gain a doorway into the way your business operates.

TL;DR

Monitor AI usage like any other business system, not as an informal staff tool. Use separate accounts, strong sign-in protection, limited integrations, and a simple response plan for unusual activity.

If usage rises when nobody is working, disconnect integrations, revoke active sessions, change credentials from a trusted device, and contact the provider promptly.

What This Means

AI services commonly keep a session active so you do not need to sign in every few minutes. That convenience can become a weakness. If malware steals a browser session, saved credential, or authorisation token, an attacker may be able to use the account without knowing your password.

An infostealer is malicious software designed to collect information stored on a device, including passwords, browser cookies, and session data. Anthropic told users that a bad actor had used infostealer malware to steal Claude login sessions and consume account usage. Source: TechCrunch

The danger is made worse when a provider shows only a total usage percentage rather than a detailed record of which device, user, task, or integration consumed it. A rising meter tells you that something happened, but not necessarily who did it, where it came from, or what information was processed.

Key insight: If an AI account is connected to business workflows, treat its login sessions and integrations as operational credentials—not as ordinary personal app access.

How This Applies to Malaysian SMEs

Many Malaysian SMEs use AI through shared email accounts, personal subscriptions, or one administrator’s login. This often starts for practical reasons: you want to test an AI writing assistant, summarise documents, prepare product descriptions, or automate replies without a lengthy technology project. The problem appears when that same account becomes essential to daily operations.

Consider a trading company that uses AI to read purchase orders arriving by email and prepare entries for its accounting system. If the AI account is compromised, an unauthorised user may consume its usage, interrupt the workflow, or view sensitive order information. Purchase orders can contain customer names, delivery addresses, product quantities, supplier details, and internal terms. Even if no funds are directly moved, the exposure can create real business disruption.

A small professional-services firm may use AI for proposals, meeting notes, website updates, and code assistance. If one employee signs in on a personal laptop that later downloads infected software, the active session could be exposed. The business may not realise anything is wrong until draft documents are altered, usage climbs unexpectedly, or the provider suspends the account for suspicious activity.

Retailers and service businesses face a different risk. You may connect AI to customer enquiries through WhatsApp, email, a website form, or a helpdesk. A compromised account could cause automated replies to fail or produce unsuitable responses. If the AI can access a knowledge base, uploaded files, or customer records, the incident may involve confidentiality as well as downtime.

The lesson is not to avoid AI. It is to decide which tasks the system can perform, what information it needs, who may authorise connections, and how you will notice abnormal behaviour. Those decisions are manageable even if you do not have a full-time IT department.

Warning Signs Worth Checking

Warning sign	What it may indicate	Immediate action
Usage rises while the team is inactive	Stolen session, unattended automation, or unknown integration	Pause connected tasks and review active sessions
Usage reaches its limit unusually quickly	Unauthorised prompts or automated workloads	Change credentials and contact the provider
New tools or integrations appear	Someone may have authorised an external service	Remove unknown connections
Unexpected sign-outs or account suspension	Provider may have detected suspicious activity	Ask for an incident review and preserve records
AI-generated replies change suddenly	Prompt, knowledge-base, or workflow tampering	Disable automation and inspect recent changes

The reported incidents included one user whose usage rose from zero to 49% within 12 minutes despite only making a few prompts and a web search. Another user reportedly used the maximum allowance every day for three days without using the account. Source: TechCrunch

Practical Takeaways for Your Business

• Assign business ownership. Keep AI accounts under a company-managed email address rather than an employee’s personal address.
• Use separate accounts by function. Do not let one shared login control customer replies, internal documents, and software development.
• Turn on multi-factor authentication. Use an authenticator app or security key where the provider supports it. Do not rely only on a password.
• Review active sessions regularly. Remove old browsers, unknown devices, former employees, and tools no longer in use.
• Limit integrations. Connect only the files, inboxes, and systems required for a specific task.
• Keep sensitive information out by default. Remove unnecessary identification numbers, passwords, bank details, and confidential attachments before sending data to an AI service.
• Use managed devices for important work. Ask staff not to sign in to business AI systems on computers used for random downloads or unverified browser extensions.
• Set a usage review routine. Check usage at least weekly and record the normal pattern for each account.
• Create an incident checklist. Include provider contact details, the person authorised to disable access, and the systems that must be disconnected.
• Keep a manual fallback. If AI supports order processing or customer replies, make sure staff can continue the essential steps without it.

A Simple Response Plan

1. Take a screenshot of the unusual usage, alerts, device list, and connected applications.
2. Pause automations and disconnect third-party tools.
3. Revoke active sessions and authorisations from a trusted, clean device.
4. Change the account password and any reused password elsewhere.
5. Check the computer for malware and update its operating system and browser.
6. Contact the AI provider and request an investigation, account protection, and usage details.
7. Review information the account could access and notify relevant customers or advisers if required.

Do not wait for perfect evidence before taking protective action. A usage spike that cannot be explained is enough reason to pause an automated workflow and investigate.

The Bigger Picture

As AI becomes part of sales, administration, finance, support, and development, account security becomes business continuity security. An AI service may look like a productivity application, but its permissions can resemble those of an employee or an integration platform.

Providers will need to offer clearer activity records, device histories, token-level reporting, better session controls, and faster support for business users. The reported case also shows why a total usage figure is not always enough: without itemised information, a customer may struggle to distinguish normal automation from theft. Source: TechCrunch

For you as an SME owner, the practical direction is clear. Start with visibility and control. Know which AI accounts exist, who owns them, what they can access, and what happens when something looks wrong. Then build safeguards around the workflows that matter most.

AI can remain useful without becoming a hidden single point of failure. The businesses that benefit most will not be those with the most tools, but those that can explain how each tool is used, detect when behaviour changes, and keep operating while an incident is being resolved.

Ready to Streamline Your Operations?

Your business should run itself. AutoRunBiz deploys AI agents to automate your daily operations — WhatsApp orders, invoicing, customer follow-ups, and accounting. Book a free 15-min ops audit to see where automation fits your business →