Can activity-monitoring software legally decide when an employee is off the clock?

The Fair Labor Standards Act uses a suffer-or-permit standard: work the employer knows about or has reason to know about is generally compensable, whether or not it was requested and whether or not a policy prohibited it. Software that deducts time when keyboard and mouse input stop is measuring input, not work — reading, thinking, being on a call, waiting for a build and handling a customer in person all look identical to idle. The deduction is a legal conclusion drawn from a proxy signal, and the employer owns that conclusion.

Whose burden is it when the automated records are wrong?

The employer is required to make and keep accurate records of hours worked. Where those records are inaccurate or incomplete, a longstanding line of authority allows an employee to carry their burden by producing sufficient evidence to show the amount of uncompensated work as a matter of just and reasonable inference, shifting the burden to the employer to negate it. Monitoring data that systematically undercounts is therefore not the defence it feels like — it can be the thing that establishes the records were unreliable.

Are auto-deducted meal breaks a problem if the AI verifies inactivity?

Automatic meal deduction has generated wage-hour litigation for years, and the recurring failure is the interrupted break: the employee who eats at their desk, answers a message, takes a call or covers a customer. Bona fide meal periods generally require the employee to be relieved from duty. Inactivity detection can show a device was unused while the employee was still working, and it cannot see work performed on another device or in person at all.

Does a policy telling employees to report unrecorded time protect us?

A clear reporting procedure that employees actually use is genuinely helpful, but it fails where the employer has reason to know work was performed and the system discouraged reporting it. If the monitoring dashboard shows activity outside recorded hours, that is constructive knowledge sitting in the employer's own database. A policy is weakest precisely where the data is strongest.

Why does automation make this a class-wide problem rather than an individual one?

Because a rule applied by software is uniform. A five-minute idle threshold, an automatic break deduction or a rounding convention is applied identically to everyone in a job classification, which supplies exactly the common policy that collective and class treatment requires. Manual timekeeping errors are idiosyncratic and hard to certify; a configuration value is a single decision affecting every member of the group.

Does the AI vendor share the liability?

The wage-hour claim runs against the employer, and in some circumstances against individuals with operational control. The vendor's exposure is usually contractual and reputational: indemnity demands, warranty claims about compliance-ready marketing, and the awkward position of having its configuration defaults quoted in a complaint. Vendors that market a product as ensuring compliance should expect that sentence to be read back to them.

AI Timekeeping and the FLSA: Where Automated Hours Become Wage-Hour Liability in 2026

Measurement Is Not the Same as the Legal Standard

Wage-and-hour law asks whether work was performed and whether the employer knew or had reason to know. Monitoring software asks whether a device registered input. Those two questions overlap enough to feel equivalent and diverge exactly where liability lives.

A support agent reading a long ticket, an engineer waiting on a deploy, a nurse charting on a shared terminal, a warehouse worker whose scanner is in someone else's hand, an employee on a phone call with a customer: all of them are working and none of them is generating keystrokes. When the system deducts that interval, it has not observed the absence of work. It has observed the absence of a proxy, and the employer has adopted the proxy as its official record.

The Recordkeeping Duty Cuts the Wrong Way

Employers subject to the FLSA must make and preserve records of hours worked. The consequence of failing that duty is not merely a technical violation: where the employer's records are inadequate, an employee may establish the extent of uncompensated work by just and reasonable inference, and the employer must then come forward with evidence to negate the inference. Damages awarded on inference are rarely favourable to the party who created the gap.

Automated systems produce enormous volumes of records, which makes it counterintuitive that they can leave an employer worse off. The volume is the point. If the underlying rule is wrong, the system generates a consistent, well-documented, easily-analysed record of the same error repeated across every employee and every pay period. Discovery in these cases is now a data export rather than a paper review, and the export favours whoever the pattern helps.

Five Configurations That Generate Claims

Idle thresholds. Any interval after which the clock stops encodes an assumption about what work looks like. The assumption is usually drawn from one job function and applied to all of them.
Automatic meal deduction. A deduction taken whether or not the break happened, in an environment where interruptions are routine and the interruption is exactly what makes the period compensable.
Rounding. Rounding conventions have to be neutral in operation, not merely neutral on their face. Combined with a system that starts the clock on first login and stops it on last input, rounding frequently drifts in one direction and the data proves it.
Boot and shutdown time. The gap between arriving and being logged into the tool that starts the clock, repeated twice daily across a large workforce, is a familiar theory in this area and automation makes the timestamps unusually precise.
After-hours signal. Message activity, mobile access and VPN sessions outside recorded shifts are constructive knowledge stored in a queryable system. Nobody has to allege the employer should have known.

The AI Layer Adds a Second Problem: Explanation

A threshold is at least legible — someone can point to the number and defend it. Newer systems infer productive time from a mix of signals: application focus, typing cadence, camera presence, task-completion telemetry, sometimes a model score. When an inferred timesheet is challenged, the employer has to explain why a specific interval was scored as non-work for a specific person on a specific day, and a blended score is a poor witness.

There is also a disparate-impact question sitting underneath the wage claim. Activity proxies systematically disadvantage employees who use assistive technology, whose duties are physical or interpersonal, who take breaks for medical or religious reasons, or who have an accommodation altering how they work. A scoring system that quietly pays those employees for fewer hours is a discrimination claim wearing a timekeeping costume, and state disclosure statutes covering electronic monitoring add notice obligations on top.

Controls That Hold Up

Never let an automated signal reduce recorded time without an affirmative human step. Flag for review; do not deduct silently.
Give employees a visible, low-friction correction path, and treat a low correction rate as a warning sign rather than a success metric.
Audit the direction of drift. Compare recorded hours against independent signals — badge, VPN, ticket timestamps — by job classification, and look at whether adjustments run one way.
Validate thresholds per role. A number derived from desk-based work does not transfer to clinical, field, warehouse or customer-facing roles.
Retain the underlying event data, not just the computed timesheet. The computation will be challenged and only the raw record can defend it.
Handle accommodations explicitly, so that an approved variation in how someone works does not become an unapproved variation in what they are paid.
Check the vendor's marketing against your own. A dashboard advertised as ensuring compliance does not transfer the obligation, and both sets of claims end up as exhibits.

Frequently Asked Questions

Our workforce is salaried. Does any of this apply?

It applies to anyone who is not properly exempt, and exemption is a fact-specific test about duties and compensation rather than a job title or a salaried payment method. Monitoring data creates a second risk here: detailed records of what an employee actually did all day are extremely useful evidence in a misclassification dispute, and they were generated voluntarily by the employer.

Do state laws add anything beyond the FLSA?

Frequently and significantly. State wage-hour regimes commonly impose daily overtime, stricter meal and rest period rules with premium pay for missed breaks, wage-statement itemisation requirements, longer limitations periods and additional penalties. Several states also require notice before electronic monitoring. The federal analysis is the floor, and the state layer is usually where the larger exposure sits.

Is remote work treated differently?

The standard is the same and the practical difficulty is greater, because the employer cannot observe the work directly and comes to rely more heavily on the proxy. Guidance in this area has generally emphasised reasonable diligence — a workable reporting process the employer does not undermine. A system that automatically strips unreported time while the employer holds data showing activity is close to the opposite of that.

Can we use the monitoring data to defend a claim?

Yes, and it is often the best evidence available — provided the methodology survives scrutiny. Ask early whether you could explain a single disputed interval to a factfinder using only what the system stored. If the answer depends on an opaque score, the data is a liability in the specific case where it matters most.

What is the exposure if the configuration has been wrong for two years?

Wage-hour claims typically reach back a period measured in years, extended where a violation is found to be wilful, and remedies commonly include unpaid wages plus an additional equal amount as liquidated damages absent a good-faith showing, together with attorneys' fees. Multiplied across a workforce, a small per-shift error becomes a material number quickly, which is why the audit is cheaper than the discovery.

We are the vendor, not the employer. What should we change?

Ship conservative defaults, make deductions require review rather than happen automatically, keep the raw event trail exportable, and be careful with compliance language in marketing. When a claim is filed against a customer, your configuration defaults and your product page are both discoverable, and neither is improved by having promised more than a timekeeping tool can deliver.

Check What Your Product Page Promises About Compliance

Workforce and HR-tech marketing makes strong claims — compliance guaranteed, accurate hours automatically, audit-ready records — that a wage-hour plaintiff will read alongside your configuration defaults.

See every compliance and accuracy claim on your site in one pass. Run a free scan and check each against what the product actually guarantees.

This article is general information and not legal advice. Wage-and-hour outcomes are highly fact-specific and state law often imposes stricter requirements than federal law. Confirm with qualified employment counsel before changing timekeeping configuration or relying on automated records.

The Software Decided They Were Not Working. The Statute Disagrees.