Digital Identity After AI Agents

From profile to proxy

A profile says something about a person.

A proxy does something for a person.

That distinction matters.

For much of the internet’s history, digital identity was built around profiles and accounts. These were ways to appear, authenticate and interact. They allowed users to enter platforms, publish content, build networks, buy services and communicate with institutions.

The profile was a digital surface.

It made a person visible.

AI agents create something different. They can become operational proxies.

They may write an email in someone’s tone. They may respond to a customer request. They may prioritise information. They may decide which message requires attention. They may prepare a legal draft, a medical summary, a travel plan or a negotiation position. They may interact with other agents, systems and organisations.

In such cases, the digital self is no longer only displayed.

It is performed.

The agent becomes part of how the person acts in the world.

This is a profound shift. It means that identity is no longer located only in what a person says about themselves or what a platform records about them. It is also expressed through automated actions taken in their name.

Delegated agency

Delegation is not new.

People have always delegated actions to others: assistants, lawyers, accountants, representatives, institutions, software tools and organisations.

But AI agents create a different type of delegation.

They are not simply instructed once. They may adapt. They may infer. They may learn preferences. They may generate responses. They may operate in uncertain environments. They may make recommendations that shape the decisions of the person they are supposed to serve.

This makes the boundary between assistance and agency less clear.

• If an AI agent drafts a message, whose voice is it?
• If it filters information, whose judgement is being applied?
• If it negotiates a price, who is responsible for the strategy?
• If it declines an invitation, reschedules a meeting or answers a request, is it merely executing an instruction or participating in social life?

The answer will depend on context, design, transparency and control.

But the underlying question remains: when a system acts on behalf of a person, how much of that action belongs to the person, and how much belongs to the system?

Digital identity after AI agents is therefore not only about authentication.

It is about attribution.

The problem of voice

Voice is central to identity.

Not only literal voice, but style, tone, judgement, rhythm, emphasis and choice of words. In digital communication, voice is how people recognise presence. It is how they decide whether something feels authentic.

AI systems can now imitate voice in both senses.

They can reproduce vocal sound. They can also reproduce written style.

This creates convenience. It also creates ambiguity.

A professional may ask an AI assistant to draft emails in their usual tone. A public figure may use AI to prepare statements. A company may automate communication while preserving the appearance of human warmth. A family may use AI to preserve or simulate the voice of someone absent.

In all these cases, the question is not only whether the output is accurate.

The question is whether the voice still belongs to the person.

• If a text sounds like someone but was generated by a system, how should it be understood?
• If a person approves an AI-generated message, is it fully theirs?
• If they do not review it, does it still carry their identity?
• If a model has learned from their past communication, who owns the style it reproduces?

These are not marginal questions. They are becoming part of ordinary digital life.

The more AI systems learn to sound like us, the more identity becomes something that can be extended, automated and imitated.

Consent and control

A digital identity shaped by AI agents requires stronger concepts of consent and control.

Traditional consent often assumes that a person agrees to a specific use of data or a specific action. AI complicates this because systems can learn from behaviour and produce future outputs that were not fully predictable at the moment of consent.

A user may consent to an assistant learning preferences.

But what exactly does that include?

An AI agent that learns from a person does not merely store information. It may build a model of preference and behaviour. That model can then shape future choices.

This means consent must become more granular, more continuous and more understandable.

Users need to know what an agent can access, what it can remember, what it can infer, what it can share, what it can do autonomously and how its permissions can be withdrawn.

Control cannot be hidden inside settings that ordinary people do not understand.

Responsibility without clarity

AI agents create responsibility problems because they distribute action across people, systems and organisations.

Imagine an AI agent sends a misleading message.

Who is responsible?

In many cases, responsibility will not sit in one place.

This is precisely the problem.

Digital systems can create action chains where intention, execution and consequence are separated. A person may set a goal. A system may choose the method. A platform may provide the infrastructure. A model may generate the content. Another system may act on the result.

This makes accountability harder.

It also makes it easier for institutions to hide behind automation.

A serious framework for AI agents must therefore ask not only what agents can do, but how responsibility is assigned when they act.

There must be clear records of delegation, clear limits of authority, clear forms of human review and clear rules for contesting harmful outcomes.

Without that, AI agents will not simply extend agency.

They will obscure it.

Trust in an agent-mediated world

Digital identity depends on trust.

People need to know who they are dealing with. They need to know whether a message comes from a person, a system, an organisation or some combination of these. They need to know when communication is automated, when content is generated and when a decision has been made by or through an AI system.

AI agents will make this harder unless transparency becomes normal.

The issue is not that every AI-assisted action must be treated as suspicious. Many forms of assistance will be ordinary and useful.

The issue is that people need to understand the status of what they encounter.

• Was this message written by a person?
• Was it drafted by AI and approved by a person?
• Was it sent autonomously by an agent?
• Is this agent authorised to speak for the person or organisation?
• What can it decide?
• What can it not decide?

Trust requires legibility.

An agent-mediated world cannot function if everyone is constantly guessing whether they are interacting with a person, a bot, a proxy, a simulation or an institution behind an interface.

The institutional dimension

AI agents will not affect only personal identity.

They will also affect institutional identity.

Companies, public services, universities, hospitals, media organisations and governments will increasingly use AI systems to communicate, classify, respond and decide.

This means that institutions will also acquire agentic layers.

A citizen may interact with an automated public service. A patient may receive information from a clinical assistant. A student may be guided by an educational AI. A customer may negotiate with a company’s automated representative.

In each case, the institution’s identity is partly expressed through an agent.

The quality, fairness and transparency of that agent become part of the institution’s character.

An irresponsible AI agent is not just a technical failure. It is an institutional failure.

This is why organisations should not treat AI agents as mere efficiency tools. They are also interfaces of trust, authority and responsibility.

The new identity contract

Digital identity after AI agents will require a new kind of contract.

Not necessarily a legal contract in the narrow sense, but a social and technical understanding of what it means for systems to act on behalf of people.

Such a contract should include at least six principles.

These principles are not obstacles to innovation.

They are conditions for trust.

Without them, AI agents may become powerful but socially unstable extensions of identity.

A different meaning of presence

AI agents also change the meaning of presence.

Until recently, being present in a digital space usually meant being connected, logged in, visible or reachable. Presence was tied to attention.

AI agents allow presence without attention.

This can be useful. It can reduce workload, extend capacity and make digital life more manageable.

But it also creates social ambiguity.

• If someone’s agent answers, have they answered?
• If an agent maintains a relationship, is the person present in that relationship?
• If professional communication is mostly agent-mediated, what happens to judgement, trust and accountability?

AI agents may therefore create a new form of distributed presence: not full absence, not full presence, but a mediated extension of the person through systems.

This is one of the defining features of being born digital now.

The self as system

The deeper transformation is that the digital self is becoming more system-like.

It is not only a profile or a collection of posts. It is an operating field made of data, permissions, models, archives, interfaces, agents and social expectations.

This does not mean the self becomes less human.

It means that the conditions through which the self appears and acts are changing.

A person’s digital identity may increasingly include:

This is not a future scenario. It is already emerging.

The challenge is to ensure that the person does not disappear inside the system built to represent and assist them.

Born digital identity

Digital identity after AI agents is not just identity online.

It is identity in action.

It is identity that can be modelled, extended, delegated, simulated and operationalised.

This creates risks: manipulation, impersonation, loss of control, blurred responsibility, institutional opacity and erosion of trust.

It also creates possibilities: assistance, accessibility, continuity, creativity, more responsive services and new forms of participation.

The future will not be decided by the technology alone.

It will be decided by design choices, social norms, laws, institutional practices and the level of digital literacy that societies are willing to build.

The central question is not whether AI agents will become part of identity.

They already are.