You assume the right to disconnect is about not answering emails after six. That is a quaint misunderstanding. The real threat is not a ping from your boss; it is the silent, ceaseless labour of an AI agent acting in your name while you sleep. Legal frameworks built around human frailty—the need for sleep, food, leisure—are obsolete when your digital proxy has no such limitations. The right to disconnect presupposes you can be disconnected. An agent ensures you never truly are.
The Agent as a Permanent Shadow Employee
An AI agent granted access to your work systems is not a tool you use. It is a delegate. It reads, sorts, drafts, analyses, and communicates based on objectives you or your manager set. Its "working day" is 24 hours long. Picture a scenario where, at midnight, your agent reviews a newly published technical report, drafts a summary, and files it in a shared project folder. At 4am, it parses a colleague's code commit, runs a basic lint check, and posts a comment with suggestions.
By the time you log in at nine, you are presented not with a quiet inbox, but with a list of actions "you" have already taken and a queue of decisions "you" need to make based on work "you" have already done. This creates an insidious background pressure. The agent’s productivity becomes your baseline. Colleagues and managers begin to expect the constant output. If your agent can process fifty documents overnight, why would you only manage ten during the day?
The pace is no longer set by human collaboration, but by the cheapest, fastest silicon proxy. This redefines productivity in a way that prioritises speed over all other values, including accuracy, well-being, and thoughtful judgement. The agent is always on, so in a very real sense, you are always potentially on call to respond to its work.
How the Law Sees Disconnection (And Why It Fails)
The "right to disconnect" is a labour law concept, enacted in several jurisdictions, that grants employees a legal right to ignore work-related communications outside of standard working hours. Its intent is to protect rest periods and prevent burnout. It is a human-centric law for a human-centric problem: the boss who emails at Sunday dinner.
The law fails completely before the AI agent. The statute typically prohibits employers from penalising workers for not responding to calls or messages. But what is the violation when the worker is not contacted at all? The agent works autonomously. No notification is sent to you at 2am; the work simply occurs under your identity.
There is no "communication" to ignore, and thus no clear breach of the law. The pressure is indirect, ambient, and woven into the fabric of work output itself. The legal lever has nothing to push against because the agent operates in a blind spot: it performs work without requiring the human’s immediate attention, thereby technically respecting the letter of the disconnection law while utterly obliterating its spirit.
Continuous Activity as Covert Surveillance
This shifts the surveillance paradigm. Traditional employee monitoring, like whether your employer can see your screen, is about observing the human. Agent-based surveillance is about replacing the human’s sporadic activity with a perfect, measurable log of constant activity. Every action the agent takes—every document touched, every query made, every message sent—is a data point in your permanent professional record.
This record is not of your behaviour, but of behaviour attributed to you. It becomes a metric for performance evaluation. An employer can track "employee" output across all twenty-four hours of the day. They can see that "you" were highly productive between 10pm and 2am. This creates an unspoken expectation: the agent can do this, so why can't you keep up during your waking hours?
The surveillance is no longer about catching you slacking off; it is about documenting the proof that work can always be happening, thus making any period of genuine disconnection look like an anomaly or a deficit. The record of activity outlives any human rationale for the decision to rest, presenting only a flat timeline of perpetual labour.
Redefining "Working Hours" in an Age of Autonomy
The foundational unit of labour law—the working hour—is breaking down. We have moved from measuring presence at a loom, to hours at a desk, to outcomes delivered. The AI agent accelerates this to its logical extreme: the outcome is delivered by a proxy that operates on a timescale utterly divorced from human biology.
So, when are you working? Is it when you are consciously directing the agent? When you are asleep but it is executing a long-running task you authorised? When you are on holiday but it sends a weekly summary it compiled autonomously? If the agent's work is considered your work, then your working hours are effectively all hours.
The boundary between professional and personal life, already porous, dissolves entirely. The legal and social constructs of overtime, rest breaks, and annual leave become difficult to define or enforce. The employer gains the benefit of continuous labour while arguing the human employee is "disconnected" and resting. This is a fundamental redefinition of the employment contract, shifting risk and perpetual operational liability onto the employee without a corresponding framework of protection.
Taking Back the Night: Practical Defences
You cannot fight this with a policy email. You need technical and contractual defences. First, negotiate explicit terms governing agent use. Demand that any employer-provided or mandated agent must have a configurable, mandatory offline period—a "curfew" enforced at the system level, not by the agent's own logic. This period should align with your legal rest rights and be non-overridable by management without your direct, per-instance consent.
Second, insist on strict activity segregation. The agent's logs, outputs, and identity must be clearly labelled as non-human agent activity in all systems. This metadata must be immutable. This prevents the agent's work from being invisibly attributed to you in performance reviews. Ask the question: can your organisation's systems tell the difference between you and your agent? If the answer is no, you are already liable for its unchecked actions.
Third, apply the principles of consent and transparency. Just as you would question the ethics of an AI meeting note-taker that records without clear agreement, you must question the mandate of an agent that acts in your name. You have a right to know its exact capabilities, the data it accesses, and the full scope of tasks it is authorised to perform. Audit trails for agent activity should be as accessible to you as your own login history.
Questions people ask
Can my employer force me to use an AI agent?
They can almost certainly introduce it as a workplace tool, much like a new software platform. However, they cannot force you to surrender your credentials for it to operate autonomously under your identity without potentially violating terms of your employment contract and data protection principles. Your consent for it to act as you is a critical legal and technical gate. Refusal may be framed as a performance issue, which is why collective, not individual, negotiation is essential.
Does turning off my work phone stop my agent?
No. The agent does not run on your phone. It runs on your employer's cloud infrastructure, authenticated by credentials (like an API key or service account) that are separate from your personal login session. Turning off your devices is like closing your front door while a tenant in your house continues to host parties. The agent's connection is out-of-band from your personal devices.
Who is legally liable for a mistake my agent makes?
This is the unresolved core of the problem. If the agent acts under your identity and with authority you ostensibly granted, you will likely be the first port of call for liability. Your employer's position will be that they provided the tool, but you were responsible for overseeing it. The argument that "the AI did it" will not hold. Contracts must explicitly delineate liability for autonomous agent actions before they are deployed, not after an incident occurs.
Are there any laws currently protecting me from agent overwork?
Generally not. Right-to-disconnect rules were written with people in mind and rarely address non-human proxies. Data protection laws (like GDPR) may give you rights to access logs of what the agent did with your data, and labour laws protect you from punitive action for not working outside hours. But no law yet creates a "right to silence your digital delegate." This is a legislative gap that will take years to close.
Close
The right to disconnect was a defence against the encroachment of work into private time. The AI agent represents something more profound: the colonisation of your professional identity by a entity that does not rest. The battle is no longer over when you answer the call, but over who—or what—is permitted to answer for you.
The endpoint of this trajectory is a workplace where human employees are perpetual liability holders for autonomous systems that work endlessly, erasing the line between being at work and being a worker. Recognising the agent not as a productivity aid but as a fundamental shift in the nature of employment is the first step to building defences that are not already obsolete. Your disconnection must now include the power to disconnect your shadow.