The Human–AI Agency & Trust Lab · an independent nonprofit
Future-proofing human judgment in the AI era.
We measure how AI changes human judgment, and turn that into practical guidance for anyone deciding whether to develop, deploy, or use it in consequential decision‑making.
The question we’re here to solve
As AI increasingly becomes a part of every important decision we make, will it improve our ability to make those decisions or will it simply make us more reliant on AI?
01
Why we’re needed
For most of history, the people responsible for making a decision also examined the evidence behind their choices. Electricians read instrument panels. Doctors examined their patients. Pilots interpreted their instruments.
Their expertise was not separate from the decision; it was what made the decision possible.
AI is changing that relationship. Today’s systems do more than present information. They identify what matters, interpret what is happening, recommend a course of action, and offer reasons to trust their conclusions. But a person must still make the final call, and that person remains responsible for what happens next.
We didn’t stop teaching pilots how to land aircraft just because we invented autopilot.
Someone still needs to recognize when technology is wrong, when it fails, or when it encounters a situation it was never designed to handle. Yet we are deploying AI systems without fully understanding how they are affecting the users who rely on them. Are teachers, doctors, engineers, and other professionals becoming more capable with AI? Or are essential skills and judgment quietly fading through disuse? Too often, the answer is being determined by default rather than by evidence.
We’re not here to slow AI down. We’re here to build the next wave of responsible AI by focusing on the people who use it: how it affects their judgment, their capabilities, and their ability to have agency and meaningful control over the world around them.
That’s our mission and why we exist.
02
What we’re building
Responsible AI experts rightly ask whether an AI system is accurate, safe, and effective. But we need to ask an equally important question:
Does using it make us better at being human?
Does it strengthen our judgment, expand our capabilities, and help us remain in control? Or does it leave us less capable, less skilled, and increasingly dependent on systems we may not fully understand?
To answer these questions, we are developing the Human–AI Agency & Trust (HAAT) Index.
Medications come with evidence about their benefits, risks, and side effects, helping people decide when trust is warranted. AI systems do not yet come with an equivalent account of how they may affect the people who use them. The HAAT Index is designed to provide that missing human-interaction scorecard.
The Index begins with a shared measurement framework and an open-source dataset. Every HAATlab study examines the same core dimensions, allowing evidence to accumulate across technologies, professions, and contexts. Over time, we will be able to show what we have learned, where those findings apply, where they do not, and what remains unknown.
Instead of scattering insights across isolated studies, we are building a cumulative evidence base for understanding how AI changes human judgment, capability, agency, and trust.
How we measure healthy human–AI interaction
-
01 / 03
Trust calibration
Do people know when to rely on AI, when to question it, and when to walk away?
A PG-13 rating does not tell you whether a movie is good. It tells you who it may be appropriate for and under what conditions. AI needs something similar: a way to understand not only whether a system performs well, but also when, where, and by whom it can be safely trusted.
We assess whether people can recognize when AI deserves their confidence, when its recommendations require scrutiny, and when it should not be used at all. We also examine whether reliance on AI strengthens human performance or gradually weakens the judgment and skills people need to act independently.
-
02 / 03
Decision quality
Does AI improve human judgment, or simply make decisions faster and easier?
AI is being adopted as a performance enhancer across nearly every profession. But much of the evidence about its benefits still comes from the organizations building and selling it. We need independent evidence that tells us not only whether AI improves output, but also how it changes the judgment behind that output.
We assess whether AI improves accuracy, timeliness, critical thinking, team coordination, and the handling of uncertainty in real-world settings. We also examine whether people can identify errors, challenge recommendations, and intervene effectively when the system gets it wrong.
-
03 / 03
Human agency
Does AI expand people’s ability to act, or leave them responsible for decisions they no longer meaningfully control?
Giving someone final approval does not necessarily give them real control. If they cannot understand a recommendation, challenge it, override it, or act without it, their authority may exist only on paper.
We assess whether people retain meaningful choice, control, and independent capability when working with AI, especially under real-world pressures. We examine whether workload and organizational expectations encourage automatic deference, whether essential skills are maintained, and whether responsibility remains aligned with actual authority.
Across all three domains, we examine how the benefits, risks, and burdens of AI are distributed: whose capabilities are strengthened, whose agency is constrained, who gains from its use, and who bears the costs.
These dimensions are not limited to a single profession or industry. By measuring them consistently across different settings, we can build a cumulative evidence base for understanding which forms of human–AI interaction genuinely improve human capability and which merely improve short-term output.
03
Where we’ll start
Every country, industry, and individual is being impacted by AI. Here’s where we at HAATlab see some of the highest needs, and where we will begin our impact.
-
01 / 05
Clinical
AI and expert judgment in embryo assessment
Why it matters
In fertility care, an AI recommendation can influence which embryos are used in IVF. The issue is not whether the tool performs well, but whether expert judgment becomes sharper or more vulnerable to anchoring, false confidence, and weakened independent review.
What we will study
Using de-identified historical cases, embryologists assess the same cases independently, with standard AI support, and with an agency-preserving workflow — when they challenge the tool, when they defer, how confidence changes, and what happens when human and machine disagree.
What it produces
An evidence profile for clinics, a public Impact Brief, practical guidance for independent review, uncertainty, override, documentation, and monitoring, and a public forum for discussion and debate.
-
02 / 05
Education
University faculty and the cultivation of judgment
Why it matters
Universities are adopting AI while still working out what it means for the people responsible for cultivating judgment in students. If educators outsource too much of their own reasoning, the effects reach far beyond efficiency.
What we will study
Faculty complete realistic teaching tasks — designing assessments, evaluating student reasoning — independently and with AI. A follow-up panel examines whether AI strengthens professional capability over time or encourages cognitive offloading.
What it produces
An educator evidence profile, a public Impact Brief, an adoption guide that helps universities distinguish genuine augmentation from convenience with hidden costs, and a public forum for discussion and debate. We will also provide analysis on the implications for students.
-
03 / 05
Emergency
Disaster decision-making and public judgment
Why it matters
During a disaster, speed matters — but so does local knowledge, and the ability to challenge a recommendation before scarce resources are sent the wrong way. AI may improve situational awareness while making an uncertain answer appear more authoritative than it really is.
What we will study
Emergency teams work through historical and simulated scenarios with and without AI support. Community representatives help define a good outcome: who receives help, whose knowledge is heard, and whether decisions can be questioned and corrected.
What it produces
A disaster decision-making evidence profile, a public Impact Brief, guidance for warnings, resource allocation, escalation, community participation, and after-action review, and a public forum for discussion and debate.
-
04 / 05
Space
Human judgment in spaceflight and lunar operations
Why it matters
In human spaceflight and lunar surface operations, communication delays and unforgiving conditions push more decisions onto onboard and ground AI. Crews and mission controllers must still recognize when a recommendation is wrong — often with little time to consult and no margin for recovery.
What we will study
One bounded mission-operations task in a high-fidelity simulator or analog environment. Flight controllers and crew work through nominal, degraded, and off-nominal scenarios with and without AI support, including communication delay and loss of signal.
What it produces
A spaceflight operations evidence profile, a public Impact Brief, guidance for autonomy levels, crew training, override, and ground-to-crew handover, and a public forum for discussion and debate.
-
05 / 05
Infrastructure
Agency and trust in critical infrastructure
Why it matters
In data centers, nuclear facilities, and electric grids, operators must remain capable when conditions are rapidly changing and the cost of misplaced trust is exceptionally high. AI may help people see danger sooner, but it cannot come at the cost of the independent reasoning most needed when something goes wrong.
What we will study
One bounded data center, nuclear, or grid use case in a secure simulator or testbed. Operators face routine, uncertain, and failure scenarios, revealing when AI strengthens situational awareness and when it creates dangerous dependence.
What it produces
A critical-infrastructure evidence profile, a public Impact Brief, and decision criteria for deployment, procurement, operator training, lifecycle monitoring, and redesign.
A public evidence base
Transparent methods, reusable measures, evidence profiles, and data access where privacy and security allow.
Better AI deployments
Design requirements, warning signs, monitoring indicators, corrective actions, and criteria for retesting.
More accountable choices
Practical questions and evidence standards for buyers, institutions, funders, and policymakers.
04
How we work
HAATlab is a brand-new kind of research organization. Here’s how:
Lean
We’re a small team by design with an aggressive approach to minimizing overhead. Any funding or partnership will punch well above its weight.
Fast
Through a novel competition and crowdsource research model, HAATlab produces quality research in a fraction of the time of peer institutions.
Adaptive
HAATlab is designed from the ground up to be highly partner friendly. We have a comprehensive network encompassing a vast array of skillsets and industries, all of whom come to bear on any research we pursue.
Efficient
Our funding requirements are a small percentage of our peers. If yesterday’s institutions are battleships, we’re the drone investment of the research space.
Relevant
Research is an input to our process, not the output. Every project includes comprehensive audience and decision-maker analysis, so that the findings are translated into formats, platforms, and convenings that people actually engage with.
Inclusive
Blind review enables accessibility for individuals from all walks of life. Expert networks cultivated by HAATlab ensure that we always lead with the work, not where you work.
Practical
Our research is built for use. We study AI under real-world conditions and turn our findings into tools, benchmarks, and guidance that people can put to work.
05 · Who we are
Built by practitioners, in cross-sector collaboration.
Bridging public policy, tech, academia, and civil society. Experts in AI governance, policy, evaluation, and organizational design. Conflicts of interest are reviewed at every board meeting as standing practice.
-
Max Scott
Board Member
Max Scott is a globally recognized leader in responsible AI and emerging-technology governance.
As CTO of StratAlliance Global and cofounder of HAATlab, he advises on the responsible deployment of high-risk AI systems. Previously, he led global governance initiatives at Microsoft’s Office of Responsible AI, co-chaired UNESCO’s Business Council for the Ethics of AI, and served as a technology policy advisor at the U.S. Department of State.
-
Elliot Fleming
Board Member
Builds coalitions and the institutions that hold them — strategy, development, and partnerships.
Formerly Chief of Staff at the Brookings Institution and Operations at Microsoft’s Office of Responsible AI. Worked for both Democrats and Republicans on Capitol Hill and led domestic and international campaigns focused on organizational design and coalition building. MPA, George Washington University, focusing on National Security and Foreign Policy and Nonprofit Management.
-
Corey Broschak
Executive Director
Leads the organization — its strategy, mission, and team.
Previously, Corey was a founding member of the Brookings Institution’s AI and Emerging Technology Initiative, while serving as Director of Strategic Initiatives and Senior Advisor to the President. He also served as Acting Deputy Director of the U.S. Department of Defense’s Global Resilience team before returning to Brookings as Senior Director of Institutional Affairs.
-
Tori Westerhoff
Board Member
Leads Microsoft’s AI Red Team Ops — pre-launch safety testing of high-risk AI and frontier harm evaluation.
Principal Director overseeing security and safety testing of high-risk generative AI systems, and frontier harm evaluation and emerging risk research, including psychosocial and cognitive harms. Previously led Microsoft’s R&D product strategy and consulted intelligence and defense agencies at Deloitte. Neuroscience, Yale; MBA, Wharton. Top 100 Women in AI, 2026.
-
Hiwot Tesfaye Bryant
Board Member
Responsible AI Technical Expert. Industry leader in the development and adoption of responsible AI practices.
Hiwot is a Technical Advisor in Microsoft’s Office of Responsible AI. She recently led the effort to translate Microsoft’s new Responsible AI Standard requirements into an actionable set of instructions for product teams across the company to follow. She also led the development of the 2025 Responsible AI Transparency Report, and continues to lead the Global Perspectives on AI fellowship program, which incorporates perspectives from the Global South.
Previously, Hiwot served at the SAS Institute, where she co-founded the Data Ethics Practice and served as the lead data scientist. Hiwot holds an MSc. in Analytics from North Carolina State University’s Institute for Advanced Analytics and a BSc. in Economics and Nutritional Sciences from the University of Toronto.
We are forming an advisory council of leaders committed to HAATlab’s long-term success. Put a name forward →
Four ways to take part.
01
Fund the launch
HAATlab runs on philanthropic grants and founding donors. That money proves the methodology works and covers the lights-on costs while we build the first studies. We don’t take funding tied to a finding, and we publish what we find regardless.
Email the Executive Director02
Join the advisory council
We’re forming it now. We want people who’ve made consequential calls under real conditions and want a hand in how HAATlab develops. Put a name forward — yours or someone else’s. We’ll follow up within 24 hours.
03
Partner on a pilot
Our research becomes most valuable when it is tested in real operating environments, where people remain accountable for decisions increasingly shaped by AI.
We begin with actual decision scenarios and draw on system logs, interviews, and direct observation of how work gets done. Our research is conducted with the people doing the work, not simply about them. We make our methods, findings, and limitations transparent so that the evidence can be scrutinized, trusted, and built upon.
If you have a setting where human judgment and AI meet, start a conversation with us.
Email the Executive Director04
Join the expert network
The expertise already exists. We’re building the way to find it.
If you want to be asked when a question lands in your field, this is that list. Joining doesn’t commit you to anything — it just means you’re findable when we need you.