Building Trustworthy AI in Government: Enablers, Guardrails, and Engagement 

    Governments are starting to use AI in areas like public services, tax work, and disaster response. When it works well, AI can help people get answers faster, spot problems earlier, and support better decisions. As a result, AI can improve productivity, responsiveness, and accountability in government.

    However, many public AI projects stay in small pilots. This happens because governments often lack skills, good data, modern digital systems, and clear ways to measure impact. These gaps can also increase risk aversion, so teams avoid innovation even when the potential benefits are high.

    The OECD proposes a simple way to understand “trustworthy AI in government”: a framework with three connected pillars. In the figure, the goal is in the centre. Around it, the three pillars explain what governments must build and do, so they can reach the public value goals shown on the outer ring (productivity, responsiveness and accountability).

     Enablers are the foundations. They include strong governance, quality data, and digital infrastructure, as well as skills and talent in the civil service. They also require purposeful investment, smart public procurement, and partnerships with non-government actors, so that AI systems can be built and used reliably.

    Guardrails are the safety systems that guide AI use. They include ethics and risk management, transparency duties, and monitoring and oversight bodies that can check results over time. They can also be non-binding guidance or binding laws and policies, along with enforcement measures. Tools like impact assessment and auditing help keep these guardrails practical. Still, guardrails should be proportionate: not every rule fits every use case, or progress may stop.

    Engagement means involving the people who are affected. This includes working across levels of government, across policy areas, and with the broader ecosystem (civil society, businesses and researchers). It also includes citizens and civil servants, and sometimes collaboration across borders. Engagement pushes governments to design user-centred systems, listen to concerns, and make necessary adjustments.

     The main message is that trust is “unlocked” by the right mix. If enablers are weak, AI cannot scale. If guardrails are missing, harms grow. If engagement is shallow, solutions may look efficient but feel unfair, and trust can fall.

(Adapted from oecd.org on February 22, 2026)

Building Trustworthy AI in Government: Enablers, Guardrails, and Engagement 

    Governments are starting to use AI in areas like public services, tax work, and disaster response. When it works well, AI can help people get answers faster, spot problems earlier, and support better decisions. As a result, AI can improve productivity, responsiveness, and accountability in government.

    However, many public AI projects stay in small pilots. This happens because governments often lack skills, good data, modern digital systems, and clear ways to measure impact. These gaps can also increase risk aversion, so teams avoid innovation even when the potential benefits are high.

    The OECD proposes a simple way to understand “trustworthy AI in government”: a framework with three connected pillars. In the figure, the goal is in the centre. Around it, the three pillars explain what governments must build and do, so they can reach the public value goals shown on the outer ring (productivity, responsiveness and accountability).

     Enablers are the foundations. They include strong governance, quality data, and digital infrastructure, as well as skills and talent in the civil service. They also require purposeful investment, smart public procurement, and partnerships with non-government actors, so that AI systems can be built and used reliably.

    Guardrails are the safety systems that guide AI use. They include ethics and risk management, transparency duties, and monitoring and oversight bodies that can check results over time. They can also be non-binding guidance or binding laws and policies, along with enforcement measures. Tools like impact assessment and auditing help keep these guardrails practical. Still, guardrails should be proportionate: not every rule fits every use case, or progress may stop.

    Engagement means involving the people who are affected. This includes working across levels of government, across policy areas, and with the broader ecosystem (civil society, businesses and researchers). It also includes citizens and civil servants, and sometimes collaboration across borders. Engagement pushes governments to design user-centred systems, listen to concerns, and make necessary adjustments.

     The main message is that trust is “unlocked” by the right mix. If enablers are weak, AI cannot scale. If guardrails are missing, harms grow. If engagement is shallow, solutions may look efficient but feel unfair, and trust can fall.

(Adapted from oecd.org on February 22, 2026)

Building Trustworthy AI in Government: Enablers, Guardrails, and Engagement 

    Governments are starting to use AI in areas like public services, tax work, and disaster response. When it works well, AI can help people get answers faster, spot problems earlier, and support better decisions. As a result, AI can improve productivity, responsiveness, and accountability in government.

    However, many public AI projects stay in small pilots. This happens because governments often lack skills, good data, modern digital systems, and clear ways to measure impact. These gaps can also increase risk aversion, so teams avoid innovation even when the potential benefits are high.

    The OECD proposes a simple way to understand “trustworthy AI in government”: a framework with three connected pillars. In the figure, the goal is in the centre. Around it, the three pillars explain what governments must build and do, so they can reach the public value goals shown on the outer ring (productivity, responsiveness and accountability).

     Enablers are the foundations. They include strong governance, quality data, and digital infrastructure, as well as skills and talent in the civil service. They also require purposeful investment, smart public procurement, and partnerships with non-government actors, so that AI systems can be built and used reliably.

    Guardrails are the safety systems that guide AI use. They include ethics and risk management, transparency duties, and monitoring and oversight bodies that can check results over time. They can also be non-binding guidance or binding laws and policies, along with enforcement measures. Tools like impact assessment and auditing help keep these guardrails practical. Still, guardrails should be proportionate: not every rule fits every use case, or progress may stop.

    Engagement means involving the people who are affected. This includes working across levels of government, across policy areas, and with the broader ecosystem (civil society, businesses and researchers). It also includes citizens and civil servants, and sometimes collaboration across borders. Engagement pushes governments to design user-centred systems, listen to concerns, and make necessary adjustments.

     The main message is that trust is “unlocked” by the right mix. If enablers are weak, AI cannot scale. If guardrails are missing, harms grow. If engagement is shallow, solutions may look efficient but feel unfair, and trust can fall.

(Adapted from oecd.org on February 22, 2026)

Building Trustworthy AI in Government: Enablers, Guardrails, and Engagement 

    Governments are starting to use AI in areas like public services, tax work, and disaster response. When it works well, AI can help people get answers faster, spot problems earlier, and support better decisions. As a result, AI can improve productivity, responsiveness, and accountability in government.

    However, many public AI projects stay in small pilots. This happens because governments often lack skills, good data, modern digital systems, and clear ways to measure impact. These gaps can also increase risk aversion, so teams avoid innovation even when the potential benefits are high.

    The OECD proposes a simple way to understand “trustworthy AI in government”: a framework with three connected pillars. In the figure, the goal is in the centre. Around it, the three pillars explain what governments must build and do, so they can reach the public value goals shown on the outer ring (productivity, responsiveness and accountability).

     Enablers are the foundations. They include strong governance, quality data, and digital infrastructure, as well as skills and talent in the civil service. They also require purposeful investment, smart public procurement, and partnerships with non-government actors, so that AI systems can be built and used reliably.

    Guardrails are the safety systems that guide AI use. They include ethics and risk management, transparency duties, and monitoring and oversight bodies that can check results over time. They can also be non-binding guidance or binding laws and policies, along with enforcement measures. Tools like impact assessment and auditing help keep these guardrails practical. Still, guardrails should be proportionate: not every rule fits every use case, or progress may stop.

    Engagement means involving the people who are affected. This includes working across levels of government, across policy areas, and with the broader ecosystem (civil society, businesses and researchers). It also includes citizens and civil servants, and sometimes collaboration across borders. Engagement pushes governments to design user-centred systems, listen to concerns, and make necessary adjustments.

     The main message is that trust is “unlocked” by the right mix. If enablers are weak, AI cannot scale. If guardrails are missing, harms grow. If engagement is shallow, solutions may look efficient but feel unfair, and trust can fall.

(Adapted from oecd.org on February 22, 2026)

Texto para questão

How do we measure attention?

    Attention, broadly defined, is the ability to direct the mind on a specific task, says Gloria Mark, author of Attention Span: A Groundbreaking Way to Restore Balance, Happiness and Productivity. There are two main types of attention, Mark explains. Involuntary attention is automatic—it’s what allows us to react to a loud noise or a jarringly bright light. Focalized attention, by contrast, is the ability to concentrate on a specific task. This latter type is what scientists measure when researching attention spans. 

    Since the early 2000s, Mark has tracked focalized attention by observing how long people remain on a task before switching to something else—such as checking email or opening a new browser tab. At first, Mark used in-person observations— researchers shadowed employees throughout the office. In recent years, she has tracked attention spans using software that monitors people’s computers.

    “Data from our first study, in 2003, revealed that people spent an average of 2.5 minutes on something before turning their attention to a different task,” she says, “Our most recent study done over the past five years shows that the figure has gone down to 40 seconds.” The measure doesn’t capture how long people can focus under ideal conditions, Mark notes, meaning shorter attention spans don’t reflect a permanent loss of attention capacity, but changes in how often people break their focus in daily life.

National Geographic. Jan 21, 2026. Adaptado.

Texto para questão

How do we measure attention?

    Attention, broadly defined, is the ability to direct the mind on a specific task, says Gloria Mark, author of Attention Span: A Groundbreaking Way to Restore Balance, Happiness and Productivity. There are two main types of attention, Mark explains. Involuntary attention is automatic—it’s what allows us to react to a loud noise or a jarringly bright light. Focalized attention, by contrast, is the ability to concentrate on a specific task. This latter type is what scientists measure when researching attention spans. 

    Since the early 2000s, Mark has tracked focalized attention by observing how long people remain on a task before switching to something else—such as checking email or opening a new browser tab. At first, Mark used in-person observations— researchers shadowed employees throughout the office. In recent years, she has tracked attention spans using software that monitors people’s computers.

    “Data from our first study, in 2003, revealed that people spent an average of 2.5 minutes on something before turning their attention to a different task,” she says, “Our most recent study done over the past five years shows that the figure has gone down to 40 seconds.” The measure doesn’t capture how long people can focus under ideal conditions, Mark notes, meaning shorter attention spans don’t reflect a permanent loss of attention capacity, but changes in how often people break their focus in daily life.

National Geographic. Jan 21, 2026. Adaptado.

Texto para questão

How do we measure attention?

    Attention, broadly defined, is the ability to direct the mind on a specific task, says Gloria Mark, author of Attention Span: A Groundbreaking Way to Restore Balance, Happiness and Productivity. There are two main types of attention, Mark explains. Involuntary attention is automatic—it’s what allows us to react to a loud noise or a jarringly bright light. Focalized attention, by contrast, is the ability to concentrate on a specific task. This latter type is what scientists measure when researching attention spans. 

    Since the early 2000s, Mark has tracked focalized attention by observing how long people remain on a task before switching to something else—such as checking email or opening a new browser tab. At first, Mark used in-person observations— researchers shadowed employees throughout the office. In recent years, she has tracked attention spans using software that monitors people’s computers.

    “Data from our first study, in 2003, revealed that people spent an average of 2.5 minutes on something before turning their attention to a different task,” she says, “Our most recent study done over the past five years shows that the figure has gone down to 40 seconds.” The measure doesn’t capture how long people can focus under ideal conditions, Mark notes, meaning shorter attention spans don’t reflect a permanent loss of attention capacity, but changes in how often people break their focus in daily life.

National Geographic. Jan 21, 2026. Adaptado.

Texto para questão

How do we measure attention?

    Attention, broadly defined, is the ability to direct the mind on a specific task, says Gloria Mark, author of Attention Span: A Groundbreaking Way to Restore Balance, Happiness and Productivity. There are two main types of attention, Mark explains. Involuntary attention is automatic—it’s what allows us to react to a loud noise or a jarringly bright light. Focalized attention, by contrast, is the ability to concentrate on a specific task. This latter type is what scientists measure when researching attention spans. 

    Since the early 2000s, Mark has tracked focalized attention by observing how long people remain on a task before switching to something else—such as checking email or opening a new browser tab. At first, Mark used in-person observations— researchers shadowed employees throughout the office. In recent years, she has tracked attention spans using software that monitors people’s computers.

    “Data from our first study, in 2003, revealed that people spent an average of 2.5 minutes on something before turning their attention to a different task,” she says, “Our most recent study done over the past five years shows that the figure has gone down to 40 seconds.” The measure doesn’t capture how long people can focus under ideal conditions, Mark notes, meaning shorter attention spans don’t reflect a permanent loss of attention capacity, but changes in how often people break their focus in daily life.

National Geographic. Jan 21, 2026. Adaptado.

Texto para questão

How do we measure attention?

    Attention, broadly defined, is the ability to direct the mind on a specific task, says Gloria Mark, author of Attention Span: A Groundbreaking Way to Restore Balance, Happiness and Productivity. There are two main types of attention, Mark explains. Involuntary attention is automatic—it’s what allows us to react to a loud noise or a jarringly bright light. Focalized attention, by contrast, is the ability to concentrate on a specific task. This latter type is what scientists measure when researching attention spans. 

    Since the early 2000s, Mark has tracked focalized attention by observing how long people remain on a task before switching to something else—such as checking email or opening a new browser tab. At first, Mark used in-person observations— researchers shadowed employees throughout the office. In recent years, she has tracked attention spans using software that monitors people’s computers.

    “Data from our first study, in 2003, revealed that people spent an average of 2.5 minutes on something before turning their attention to a different task,” she says, “Our most recent study done over the past five years shows that the figure has gone down to 40 seconds.” The measure doesn’t capture how long people can focus under ideal conditions, Mark notes, meaning shorter attention spans don’t reflect a permanent loss of attention capacity, but changes in how often people break their focus in daily life.

National Geographic. Jan 21, 2026. Adaptado.

Texto para questão

How do we measure attention?

    Attention, broadly defined, is the ability to direct the mind on a specific task, says Gloria Mark, author of Attention Span: A Groundbreaking Way to Restore Balance, Happiness and Productivity. There are two main types of attention, Mark explains. Involuntary attention is automatic—it’s what allows us to react to a loud noise or a jarringly bright light. Focalized attention, by contrast, is the ability to concentrate on a specific task. This latter type is what scientists measure when researching attention spans. 

    Since the early 2000s, Mark has tracked focalized attention by observing how long people remain on a task before switching to something else—such as checking email or opening a new browser tab. At first, Mark used in-person observations— researchers shadowed employees throughout the office. In recent years, she has tracked attention spans using software that monitors people’s computers.

    “Data from our first study, in 2003, revealed that people spent an average of 2.5 minutes on something before turning their attention to a different task,” she says, “Our most recent study done over the past five years shows that the figure has gone down to 40 seconds.” The measure doesn’t capture how long people can focus under ideal conditions, Mark notes, meaning shorter attention spans don’t reflect a permanent loss of attention capacity, but changes in how often people break their focus in daily life.

National Geographic. Jan 21, 2026. Adaptado.