You have probably felt it already: you call your bank or health insurer, and instead of a person you get a cheery synthetic voice that does not quite understand your problem—especially when it is urgent, scary, or messy.

For simple stuff, automated systems are fine. You might even prefer them. But when the stakes are high—your mortgage application, your disability benefits, your medical treatment—being stuck in an endless bot loop can feel less like “innovation” and more like being locked out of your own life.

That tension is exactly what policy-makers are now grappling with. Behind the jargon of “automated decision-making,” “high‑risk AI,” and “human oversight,” there is a basic question that matters to you: in essential services, do you still have a right to human contact?

In this post, we will unpack what that right actually looks like in law today, how new AI rules are raising the bar, and what it means for you when tools like ChatGPT, Claude, and Gemini start to sit between you and your most critical services.

What do we mean by “essential services”?

Before we talk rights, we need to talk scope. Not every annoying chatbot encounter is a human‑rights issue.

When regulators and legal scholars talk about essential services, they typically mean things that directly affect your basic needs or legal status, like:

  • Banking and credit (e.g., getting a loan, opening an account)
  • Insurance and pensions
  • Healthcare and mental health support
  • Education and exams
  • Core utilities (electricity, water, internet)
  • Employment decisions (hiring, firing, promotions)
  • Government services and benefits (visas, welfare, disability, housing)

These are precisely the areas where the EU’s AI Act treats many systems as “high‑risk” and requires strong human oversight, record‑keeping and transparency, because they can affect fundamental rights and opportunities in your life.Regulation of artificial intelligence overview

So while your supermarket’s Black Friday chatbot might just be annoying, an AI that decides whether you get a cancer treatment or a student loan is in a very different category.

If you have heard anything about a “right to a human” in AI, it probably comes from Article 22 of the EU General Data Protection Regulation (GDPR).

This provision limits “solely automated” decisions that produce legal or similarly significant effects on you—classic examples include:

  • Credit scoring that determines whether you get a loan
  • Automated hiring systems that reject your application
  • Insurance pricing that dramatically changes your premium

Under GDPR, these kinds of decisions are generally not allowed if they are made purely by machines, unless specific exceptions apply (like explicit consent or a law allowing it). And even when they are allowed, the controller must put safeguards in place—most importantly, your right to:

  • Obtain human intervention
  • Express your point of view
  • Contest the decision

That is spelled out in the text and in the EU’s own guidance on automated decision‑making and profiling.GDPR Article 22 textArticle 29 Working Party guidance

So while GDPR does not literally say “you always have a right to talk to a human,” it does say: if an important decision about you is being made by a machine alone, you must be able to get a human involved and challenge the outcome.

That is already a powerful, legally recognized limit on fully automated AI in key services.

The EU AI Act raises the stakes for high‑risk systems

Fast‑forward to the EU AI Act, a dedicated AI regulation that came into force in 2024 and starts applying in stages over the next few years. It uses a risk‑based approach:

  • Some uses of AI are prohibited (like certain social scoring and real‑time biometric identification in public spaces).
  • Many systems in essential services are classified as high‑risk and face strict requirements.
  • Lower‑risk systems face transparency obligations (for example, telling you when you are talking to a bot).

For high‑risk AI, such as systems used in credit scoring, hiring, education, health, and access to essential services, providers must build in:

  • Clear information and instructions for use
  • Comprehensive logging and record‑keeping
  • Robust risk management and data governance
  • Crucially, effective human oversight so that people can “overrule” or intervene in AI outputs

Overviews of the AI Act emphasize that high‑risk systems must not be left to run on autopilot; they need real human control mechanisms, especially where rights and livelihoods are involved.EU AI Act risk classification summaryArtificial Intelligence Act overview

In plain language: if your bank, hospital, or public authority is using an AI system that falls into these high‑risk categories, they are not supposed to say “sorry, the algorithm decided, nothing we can do.” The law expects a meaningful human safety net.

The Italian ChatGPT pause: a warning shot

This is not just theory. In March 2023, Italy’s data protection authority (the Garante) made global headlines when it ordered ChatGPT temporarily blocked in Italy over GDPR concerns, including transparency, data accuracy, and lack of effective age checks.Ars Technica coverage of Italian ban

The case was about a general‑purpose chatbot rather than an essential‑services decision engine. But it illustrated two key trends:

  1. Regulators are willing to intervene aggressively when they think AI systems are deployed without adequate safeguards for vulnerable users (like children).
  2. General‑purpose models such as ChatGPT, Claude, and Gemini are increasingly being integrated into banking, health, and government workflows—so the line between “generic chatbot” and “critical service interface” is blurring fast.

After OpenAI implemented changes around transparency and age‑gating, ChatGPT access was restored in Italy. But the episode is now frequently cited as a case study in how authorities might act when AI tools sit directly between the public and sensitive services.WilmerHale analysis of the Garante decision

For you as a user, the takeaway is clear: regulators are increasingly expecting that when AI is used in important contexts, it must be coupled with clear information, safeguards, and human fallback.

Human oversight vs real human contact

Here is where things get subtle—and where service providers sometimes exploit the ambiguity.

Legal concepts like:

  • “The right to obtain human intervention” (GDPR)
  • “Human oversight” (EU AI Act and other frameworks)
  • “User‑centric design” and “explainability” (various guidelines)

do not automatically translate into “you can always talk to a human whenever you want.”

In practice:

  • Human oversight can mean an employee reviewing batches of AI decisions after the fact.
  • A “human in the loop” might be someone briefly checking an AI score on a screen and clicking “approve” 100 times an hour.
  • “Human intervention” could be offered only through complex complaint processes that take weeks.

From a legal risk perspective, that might be just enough to claim compliance. From your perspective—especially if you have just been denied a loan or wrongly flagged for fraud—that can feel like no real human contact at all.

This is why some scholars and advocacy groups argue for much clearer, enforceable rights to direct access to a human decision‑maker in essential services, not just abstract “oversight” somewhere in the background.

Where current law quietly limits pure AI in essentials

Even without a neat, one‑sentence “right to human contact” law, there are already several built‑in constraints that matter in practice:

  1. Prohibition on purely automated, high‑impact decisions (GDPR Article 22)
    If a decision has legal or similarly significant effects and is made solely by automated means, you generally must have the ability to get a human review and contest it. That inherently limits “AI‑only” decision pipelines in critical areas.

  2. High‑risk classification in the EU AI Act
    When AI is used in access to essential services, the Act demands meaningful human oversight, documentation, and risk management. Supervisory authorities can demand changes or impose penalties if systems effectively shut people out from real recourse.

  3. Sector‑specific rules
    In banking, insurance, health, and employment, there are often existing laws requiring fair treatment, non‑discrimination, and explainability of decisions, which are hard to meet with black‑box, fully automated systems. Compliance regimes are increasingly reading those obligations together with AI‑specific rules.

  4. Emerging national clarifications
    Some jurisdictions (for example, in discussions around a revised UK GDPR) are explicitly restating the idea that people affected by automated decisions should be able to contest them and seek human intervention in plain language.EU analysis of UK GDPR reforms

None of this guarantees a friendly human on the first ring. But the legal trend line is clear: the more essential the service, the less acceptable it is to hide entirely behind an algorithm.

What this means for people building with AI

If you are on the product or engineering side, this has real design implications for how you integrate tools like ChatGPT, Claude, or Gemini into critical workflows.

To avoid crossing the line into “unlawful automation,” you should be thinking about:

  • Clear scoping: Do not let a chatbot silently creep from answering FAQs into making binding eligibility decisions.
  • Visible escalation paths: Make it easy and obvious for users to request a human when issues are complex, sensitive, or contested.
  • Documented human roles: Spell out when humans review AI outputs, what they can change, and how they are trained to avoid rubber‑stamping.
  • Logging and auditability: Keep records of both machine suggestions and human overrides to show regulators that oversight is real, not just theoretical.
  • Accessible explanations: If an AI‑assisted decision denies someone a service, give them a way to understand why in plain language and how to appeal.

You can still use powerful models to speed up work and improve consistency—but in essential services, the human has to stay in the loop in a way that is genuine, not decorative.

So do you have a “right to human contact” yet?

In most places, you do not have a simple, across‑the‑board legal right to human contact for every AI interaction.

What you increasingly do have, especially in Europe and in services that deeply affect your life, is:

  • A right not to be subject solely to automated decisions with serious impacts, unless strict conditions are met.
  • A right to demand human intervention and to contest such decisions when they are allowed.
  • Growing expectations that AI used in essential services will be subject to human oversight, transparency, and accountability.

That is not the same as banning AI from essentials. It is more like saying: you can absolutely use AI here—but you still owe people real humans with real responsibility when it truly matters.

How you can protect yourself and push for better systems

Here are concrete steps you can take, whether you are a user or someone working in an essential‑service organization:

  1. Ask explicitly:

    • “Is this decision being made solely by an automated system?”
    • “How can I request a human review?”
      In EU contexts, referencing “my rights under GDPR Article 22” can quickly move you out of the bot maze.

  2. Look for escalation options:
    When you are stuck in an AI loop, persist in finding buttons or prompts like “talk to an agent,” “file a complaint,” or “request review.” Document your attempts (screenshots, call logs) in case you need to escalate to an ombudsman or regulator.

  3. Inside organizations, design for dignity:
    If you are building or buying AI for essential services, bake in simple, prominent human fallback from day one. Treat “Can the user reach a responsible human quickly if this goes wrong?” as a non‑negotiable design requirement, not a nice‑to‑have.

The march toward AI‑mediated services is not going to stop. But we do not have to accept a future where a model has the final word on your health, your money, or your freedom. The law is already sketching out a right to meaningful human involvement; the next step is to turn that into an everyday reality whenever you pick up the phone and ask for help.