Require human-in-the-loop oversight for agentic AI systems acting in high-stakes domains

ai-safety ai-governance policy ethics ai-regulation regulations

Share this with your representatives, journalists and friends

Key insights

Across supportive quotes, oversight is usually tied to accountability, auditability, transparency, monitoring, and redress rather than treated as a stand-alone requirement.
Several quotes define meaningful oversight as more than review: the human must have authority, competence, time, independence, and the power to alter or stop outcomes.
Supporters most often invoke oversight where AI can affect life, safety, rights, public services, finance, education, healthcare, employment, or critical infrastructure.
Opposing quotes do not argue for leaving agents uncontrolled; they question blanket or pervasive human approval requirements and emphasize faster, more embedded containment and governance.
The quotes show an internal design debate even among supporters, with some favoring explicit approval gates and others favoring graduated models such as human-on-the-loop or human-in-command.

Arguments for

Human final authority

Proponents say decisions that can cost a life, livelihood, or fundamental right, or materially affect public services and other significant interests, should not be delegated to AI alone. In these settings, humans should retain the final word, especially for lethal force and other decisions affecting rights or eligibility.

However capable of our systems become, a person must keep the final word to understand the decision, to contest it, and where it matters most, to switch the systems off. This is non-negotiable, above all in high-risk systems where a single decision can cost a life, a livelihood, …

Rafal Kownacki · Jul 7, 2026 · full quote
Humans must retain final control over decisions supported or affected by AI. This is particularly important when AI systems can affect access to rights, public services, employment opportunities, education, healthcare, credit, information, or democratic participation.

Eugênio Garcia · Jul 7, 2026 · full quote
No machine should ever be given the power to decide to kill a human being on its own.

Don Beyer · Jul 17, 2026 · full quote
Where AI affects fundamental rights or the delivery of public services, meaningful human oversight, transparency and effective mechanisms for accountability and redress must remain central.

Ntumeng Majara · Jul 6, 2026 · full quote
AI should not serve as the sole basis for decisions that materially affect a student’s academic standing, progression, financial status, enrollment status, access to university services, disability accommodations, student conduct outcomes, or other significant educational interes…

Aspen University · Jul 24, 2026 · full quote

Real accountability and redress

Supporters argue that oversight keeps responsibility with identifiable people and institutions rather than allowing anyone to say the system made the decision. They call for named humans with authority, competence, time, independence, and power to change outcomes or stop systems, along with contestability and timely redress.

Human oversight of AI systems cannot be just a rubber stamp. It requires that an identified person be granted authority, competence, time, independence, and power to alter and even stop a system.

Volker Türk · Jul 7, 2026 · full quote
We need to appoint an identifiable human being in charge for all issues that are tied to rights, and there needs to be a right to redress in good time, in line with the speed of AI systems, so that humans can receive redress if necessary. Therefore, the functioning of the algorit…

Amal El Fallah Seghrouchni · Jul 7, 2026 · full quote
The true challenge is to ensure that its use is transparent, responsible, can be audited, and completely compatible with ethical judicial priorities, human rights, and the rule of law. No public institution can justify a decision by saying that that is what the system told us to …

Samuel Arsenault · Jul 7, 2026 · full quote
Those responsible for the different phases of the AI system lifecycle should be identifiable and accountable for the outcomes of the AI systems, and human oversight of AI systems should be enabled.

Australia’s Data and Digital Ministers · Aug 25, 2026 · full quote
Human oversight must be real, informed, and able to change outcomes.

UNHCR · Jul 7, 2026 · full quote

Built-in control mechanisms

Another argument is that oversight should be designed into deployment through approval workflows, escalation paths, audit trails, monitoring, and independent shutdown or kill-switch capabilities. High-risk actions should be held at gates for human approval, and systems should remain stoppable even without the AI's cooperation.

Implement accessible kill switches, override controls or independent shutdown mechanisms for every autonomous or safety-relevant edge AI system. These controls should not rely on the AI’s cooperation and should be tested regularly. Define where human approval is required before A…

Canadian Centre for Cyber Security · Jul 15, 2026 · full quote
Monitor agent actions. Use auto-review mode to review tool calls outside the Codex sandbox before they execute. Add further monitoring and human oversight for higher-risk workflows.

OpenAI · Aug 10, 2026 · full quote
An "audit trail should preserve both identities: the human who initiated or authorized the action and the agent that executed it."

Vinay Patel · Jul 29, 2026 · full quote
Make the technical controls real. That means AI-ready data, clear access controls, audit trails, model and agent monitoring, defined escalation paths and the ability to intervene when something behaves outside expectations.

Joe Depa · Aug 7, 2026 · full quote
Advanced AI models should never be deployed without a reliable off switch. The AI Kill Switch Act establishes a commonsense safeguard by requiring leading AI companies to maintain the ability to shut down their models and empowering the federal government to act when a deployed s…

Brad Carson · Jul 23, 2026 · full quote
Design for human oversight from Day 1: Don't bolt on governance after deployment. Build approval workflows, block states, and audit trails directly into your architecture.

Jaleh Reeves · Jul 1, 2026 · full quote

AI as assisted judgment

A further supportive line holds that agentic AI works best as decision support rather than a replacement for human judgment. AI may read more data or execute more routine work, but people still need domain expertise to frame questions, evaluate outputs, validate results, preserve skills, and make final decisions.

In writing my forthcoming book, The Trust Code, I interviewed the people who run AI inside the world’s most critical systems, from energy grids and payment networks to technology giants. One pattern kept surfacing. The organizations that scale AI with confidence are the ones that…

Tiffany Xingyu Wang · Sep 1, 2026 · full quote
We conclude that effective use of agentic AI requires both domain expertise to frame questions and evaluate outputs, and coding literacy to guide the AI's approach.

Edmund Seto · Aug 7, 2026 · full quote
Ensuring meaningful human oversight of AI systems through various types of checks will ensure that the employees keep their core skills sharp for the task, but allowing the shift of work into new tasks will allow the building of new skills that were not possible prior to the over…

Janet Bastiman · Aug 12, 2026 · full quote
The leaders who get this right treat AI as an extra pair of eyes. The ‘digital twin’ of AI technology reads far more EHS data than any human team could but still leaves the decision to a person.

Tom Goodmanson · Jul 15, 2026 · full quote
A health worker trained to trust an AI system they can’t interrogate is not empowered, leading to mistakes outside their control.

Hans Henri P. Kluge · Jul 15, 2026 · full quote

Arguments against

Human bottlenecks at scale

Critics argue that requiring humans to stay in the loop for too many agent actions becomes unrealistic in fast production environments with many agents and decisions. In this view, human-in-the-loop should be used only where absolutely necessary, with more sophisticated governance and accountability frameworks doing the rest.

Our frameworks were not built to contemplate autonomous agents, and relying on a human in the loop for all agent actions is unlikely to be realistic. More sophisticated governance and accountability frameworks may be needed.

Sarah Breeden · Jul 2, 2026 · full quote
We’re not huge fans of human-in-the-loop. It’s something that you should use judiciously, where you absolutely need it. But it’s not something that you can do at high velocity. You will not get the results that you want to get.

Eric Brandwine · Jun 20, 2026 · full quote
The industry’s current answer is unsatisfying: Human-in-the-Loop (HITL). In development environments and low-frequency pipelines, routing uncertain decisions to a human can be defensible. In production systems operating at scale—dozens of agents, hundreds of decisions per hour—it…

Artur Huk · May 11, 2026 · full quote

Automated response for security

Another opposing line says security controls sometimes need to react immediately when an agent behaves strangely or trust changes, rather than waiting for human approval or a ticket. Proponents of this view favor continuous verification and automatic restriction of access to prevent a larger incident.

A lot of tools can tell you an AI agent is behaving strangely. Very few can act on that at the network layer, without waiting on a human to approve a ticket. That’s the piece we’re closing.

Garrett Gross · Aug 18, 2026 · full quote
Every identity that can connect, access data, or act must be continuously verified and governed. Portnox gives organizations the ability to immediately restrict access when trust changes, without waiting for an AI agent to create a larger security incident.

Denny LeCompte · Aug 18, 2026 · full quote

Middle ground

Risk-based oversight models

Several quotes support oversight but reject a single blanket pattern. They argue that the right model depends on risk, confidence, ambiguity, regulatory sensitivity, operational maturity, autonomy, and consequence, so humans may be in the loop, on the loop, or in command at different points rather than at every level.

Human-in-the-loop should be designed as a graduated oversight model rather than a single pattern. The right mode depends on risk, confidence, ambiguity, regulatory sensitivity, operational maturity, and the trustworthiness of the automation.

Florian Lauck-Wunderlich · Jul 24, 2026 · full quote
To protect employees and exact real productivity, businesses must build in human oversight at the right level – not at every level.

Sonali Fenner · Jul 20, 2026 · full quote
Human oversight should increase as agent autonomy and consequence increase.

PwC · Jul 17, 2026 · full quote
Our national AI policy ensures human in the loop where AI must support and not replace human decision-making. Human on the loop, where human oversight remains an inbuilt feature. And human in command, where ultimate legal responsibility remains that of humans.

Shahza Fatima Khawaja · Jul 7, 2026 · full quote
Conventional oversight models are increasingly insufficient. The shift is moving from “human-in-the-loop” to “human-on-the-loop”, reflecting a more supervisory rather than directly interventionist role for people.

Pieter Danhieux · Jul 13, 2026 · full quote

Oversight capacity limits

One explicitly undecided view says existing governance frameworks assume competent human overseers will remain present and capable of exercising meaningful control. It therefore argues that accountability should be embedded as a capacity-sensitive property of system architecture, not depend only on human presence.

Existing responsible AI governance frameworks mandate human oversight. Yet they share a critical assumption: that competent human overseers remain present and capable of exercising meaningful oversight and control over AI decisions. Accountability must be treated as a capacity-se…

Gabriel Sze · Apr 20, 2026 · full quote

The quotes show broad support for keeping agentic AI under human control in consequential settings, but they differ on what form that control should take in practice. Supporters emphasize final human authority, accountability, and built-in safeguards, while skeptics question whether direct human approval can remain meaningful at operational speed and scale. The live issue in the debate is therefore the design of oversight, not only its desirability.

Cast your vote:

Results (538 votes):

Total (538 votes)

For 522 (97%) Abstain 1 (0%) Against 15 (3%)

For (518)

Fareed Zakaria
CNN host, Washington Post columnist, author, and political commentator.

I am this person! Add a sourced quote Report comment

votes For and says:

This is a systemic problem that we cannot leave to the good graces of private companies. When thinking about regulations, we should focus centrally on how much autonomy we give these systems. A chatbot that answers a question poses one set of risks. ...
more AI Verified source (Sep 18, 2026)

added 2d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Alastair Denniston
NHS consultant and Chair of the National Commission into the Regulation of AI in Healthcare.

I am this person! Add a sourced quote Report comment

votes For and says:

One message came through clearly: people are open to AI improving their care, but only if it is safe, overseen by humans, and if they know when it is being used. These recommendations are designed to make that trust possible, so that the UK can lead ...
more AI Verified source (Sep 10, 2026)

added 9d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Steven Mills
Chief AI Ethics Officer and Managing Director & Partner at BCG

I am this person! Add a sourced quote Report comment

votes For and says:

Deeper human review is concentrated on the AI products that carry the greatest risk.
AI Verified source (Sep 10, 2026)

added 10d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Menlo Security
Cybersecurity company providing browser and AI-agent security products for enterprises.

I am this person! Add a sourced quote Report comment

votes For and says:

At that point, speed matters less than judgment (aka security). A person needs to understand the context, weigh the risk, and decide whether the action should proceed. As agents take on more routine execution, human involvement shifts toward these le...
more AI Verified source (Sep 8, 2026)

added 7d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Australian Government Department of Finance

I am this person! Add a sourced quote Report comment

votes For and says:

Governments should provide clear, simple explanations for how an AI system reaches an outcome. This includes: [...] the implementation of human oversight. [...] When an AI system influences or is used as part of administrative decision making, decisi...
more AI Verified source (Sep 8, 2026)

added 12d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Mohan Veloo
Chief Technology Officer for Asia Pacific, China and Japan at F5, specializing in enterprise technology and security.

I am this person! Add a sourced quote Report comment

votes For and says:

Human approval should remain mandatory for actions that could jeopardize human safety, create material financial loss, change production systems, delete or disclose sensitive data, alter security policies, create legal commitments, affect employment ...
more AI Verified source (Sep 7, 2026)

added 13d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Basile Njei
Associate Professor Adjunct at Yale School of Medicine who studies AI in clinical care.

I am this person! Add a sourced quote Report comment

votes For and says:

The human being is going to be the orchestrator.
AI Verified source (Sep 1, 2026)

added 14d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Tiffany Xingyu Wang
CEO of Songsheet; writer on AI governance, trust, and human judgment in automated decision-making.

I am this person! Add a sourced quote Report comment

votes For and says:

In writing my forthcoming book, The Trust Code, I interviewed the people who run AI inside the world’s most critical systems, from energy grids and payment networks to technology giants. One pattern kept surfacing. The organizations that scale AI wit...
more AI Verified source (Sep 1, 2026)

added 17d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Australia’s Data and Digital Ministers
Australian federal, state, and territory ministers responsible for data and digital policy.

I am this person! Add a sourced quote Report comment

votes For and says:

Those responsible for the different phases of the AI system lifecycle should be identifiable and accountable for the outcomes of the AI systems, and human oversight of AI systems should be enabled.
AI Verified source (Aug 25, 2026)

added 27d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
OpenAI
AI research organization

I am this person! Add a sourced quote Report comment

votes For and says:

Guided by our National Security Principles, we approach national security deployments with safeguards, meaningful human judgment, and appropriate oversight, especially in high-stakes settings.
AI Verified source (Aug 18, 2026) 1 of 3

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Cathal McCarthy
Chief Strategy Officer at Kore.ai, writing on AI governance and enterprise AI.

I am this person! Add a sourced quote Report comment

votes For and says:

Organizations need processes to monitor, challenge and review AI systems, particularly as they become more autonomous.
AI Verified source (Aug 14, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Janet Bastiman
Chief data scientist at Napier AI, writing on responsible AI, human oversight, and workforce skills.

I am this person! Add a sourced quote Report comment

votes For and says:

Ensuring meaningful human oversight of AI systems through various types of checks will ensure that the employees keep their core skills sharp for the task, but allowing the shift of work into new tasks will allow the building of new skills that were ...
more AI Verified source (Aug 12, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Henry Han
Researcher on AI governance, verifiability, and auditability in financial technology.

I am this person! Add a sourced quote Report comment

votes For and says:

Financial institutions are delegating consequential decisions to agentic AI systems that decompose goals, coordinate models and tools, and act with little oversight. Yet agentic AI governance in FinTech is under-investigated.
AI Verified source (Aug 11, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Amanda Caswell
AI Editor at Tom's Guide and technology journalist.

I am this person! Add a sourced quote Report comment

votes For and says:

Human-in-the-Loop gates: High-stakes actions like sending emails, modifying files, executing code, or moving money, require explicit human confirmation before the AI can proceed.
AI Verified source (Aug 10, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Ricardo Britto
Author of a 2026 arXiv paper on assurance for AI-native software development.

I am this person! Add a sourced quote Report comment

votes For and says:

The AI-Native Manifesto envisions large-scale agile software development in which humans increasingly govern intent, risk, and exceptions while agents execute more of the engineering process.
AI Verified source (Aug 7, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Edmund Seto
Environmental health researcher and author of a 2026 paper on human-in-the-loop agentic AI.

I am this person! Add a sourced quote Report comment

votes For and says:

We conclude that effective use of agentic AI requires both domain expertise to frame questions and evaluate outputs, and coding literacy to guide the AI's approach.
AI Verified source (Aug 7, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Joe Depa

I am this person! Add a sourced quote Report comment

votes For and says:

Make the technical controls real. That means AI-ready data, clear access controls, audit trails, model and agent monitoring, defined escalation paths and the ability to intervene when something behaves outside expectations.
AI Verified source (Aug 7, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Vishal Salvi

I am this person! Add a sourced quote Report comment

votes For and says:

Organizations should also define where human intervention is required and who is responsible for the decisions models make. Oversight should focus on the activities that carry the greatest operational, financial or regulatory risk.
AI Verified source (Aug 7, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Rodrigo Coelho
Chief Executive Officer of Edge & Node, writing on AI accountability and security.

I am this person! Add a sourced quote Report comment

votes For and says:

Embedding infrastructural policy enforcement, audit trails, and human-reviewable records does bring an additional layer prior to deployment. And in a domain where nefarious actors operate free of this operational layer, it's fair to ask whether enter...
more AI Verified source (Aug 6, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Daniel Card

I am this person! Add a sourced quote Report comment

votes For and says:

Security comes from the details; it comes from care. More care needs to go into systems – even more so when they are autonomous. We can't just set them off and hope they are ok. They need monitoring.
AI Verified source (Aug 5, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Nathan DeBardeleben

I am this person! Add a sourced quote Report comment

votes For and says:

Its constructive proposal is engineered heterogeneity: the process that proposes an action should not serve as its sole approver and auditor. Independent monitoring and review over time provide additional checks on that process.
AI Verified source (Aug 4, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Branka Marijan
Senior fellow at the Centre for International Governance Innovation and senior researcher at Project Ploughshares.

I am this person! Add a sourced quote Report comment

votes For and says:

Governance must ensure that commanders have sufficient training, time and authority to critically evaluate AI recommendations, and that human decision makers retain careful, independent judgment.
AI Verified source (Aug 4, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Vinay Patel
Chief Security Officer at Zendesk, specializing in enterprise and agentic-AI security.

I am this person! Add a sourced quote Report comment

votes For and says:

An "audit trail should preserve both identities: the human who initiated or authorized the action and the agent that executed it."
AI Verified source (Jul 29, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Edward Chen
Chief AI Officer at NCS

I am this person! Add a sourced quote Report comment

votes For and says:

Organisations should not govern agents like software modules. They should onboard them like digital workers – with clear roles, access limits, supervision and accountability.
AI Verified source (Jul 28, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Hugh Scantlebury

I am this person! Add a sourced quote Report comment

votes For and says:

While AI is becoming an increasingly trusted part of the workplace, human oversight and validation remain essential.
AI Verified source (Jul 27, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Ceren Kaya Akgün
Author and practitioner writing on human-in-the-loop controls for agentic AI workflows.

I am this person! Add a sourced quote Report comment

votes For and says:

Human in the loop for AI agents is an execution-control pattern in which an agent stops before a designated action, hands a decision to a person, and does not continue until that person accepts, edits, or refuses it.
AI Verified source (Jul 25, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Florian Lauck-Wunderlich
Enterprise AI practitioner writing on agentic-AI governance and human-in-the-loop architecture at Pega.

I am this person! Add a sourced quote Report comment

votes For and says:

Human-in-the-loop should be designed as a graduated oversight model rather than a single pattern. The right mode depends on risk, confidence, ambiguity, regulatory sensitivity, operational maturity, and the trustworthiness of the automation.
AI Verified source (Jul 24, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Aspen University
Accredited U.S. university and publicly traded higher-education institution.

I am this person! Add a sourced quote Report comment

votes For and says:

AI should not serve as the sole basis for decisions that materially affect a student’s academic standing, progression, financial status, enrollment status, access to university services, disability accommodations, student conduct outcomes, or other s...
more AI Verified source (Jul 24, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate

Loading more opinions...

Abstain (1)

Loading more opinions...

Against (15)

Garrett Gross
Field Chief Information Security Officer at Portnox, an enterprise access-control provider.

I am this person! Add a sourced quote Report comment

votes Against and says:

A lot of tools can tell you an AI agent is behaving strangely. Very few can act on that at the network layer, without waiting on a human to approve a ticket. That’s the piece we’re closing.
AI Verified source (Aug 18, 2026)

added 29d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Denny LeCompte
Chief Executive Officer of Portnox, a cloud-native enterprise access-control provider.

I am this person! Add a sourced quote Report comment

votes Against and says:

Every identity that can connect, access data, or act must be continuously verified and governed. Portnox gives organizations the ability to immediately restrict access when trust changes, without waiting for an AI agent to create a larger security in...
more AI Verified source (Aug 18, 2026)

added 29d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate

Loading more opinions...

More h/ai-regulation votes

Ban AI companion chatbots for minors under 18

Ban autonomous lethal weapons

AI companies should be liable for harms caused by their deployed models

AI agents should be granted legal personhood as non-human corporations

Require all advanced AI chips to be registered and traceable to their physical location

Back