Cognitive Surrender: AI Trust Risks for Nonprofits
"48% higher on practice problems. 17% lower on the exam." That's what happened to high school students who used AI on math practice, then had to sit a test without it, according to a study published in PNAS. It looked like progress. It was debt.
A new review in Trends in Cognitive Sciences, asks the question directly: is AI making us stupid? The findings are nuanced.
Our foundational abilities, like working memory and attention, appear resilient to AI. What erodes is specific skills, when we skip the grind and the effortful thinking that build them.
One of the most insidious findings in the review has a name: cognitive surrender. Not using AI as a tool, but handing over judgment entirely, and adopting its answer as your own without checking.
That's a psychology finding. But inside an organization, it becomes an operational risk. Here's what it looks like, and what to do about it.
Three Key Organizational Risks of Cognitive Surrender
Surrender is invisible until something breaks
In the math study, students didn't feel themselves accumulating debt. Their practice scores climbed. Everything looked fine, until the exam came without AI, and the debt came due.
Inside an organization, the exam is a wrong AI-drafted eligibility decision that goes out the door. Or a client quote misrepresented in a donor email nobody reread. By the time you notice, the gap has been open for months. Nothing in the dashboards flagged it, because the work still got done. It just got done wrong, quietly, at a pace that looked like productivity.
E.g. A program team using AI to summarize client intake forms for months, catching nothing, until a summary drops a disclosed accessibility need and the client shows up to a session the organization wasn't prepared for.2. Surrender concentrates where accountability is already thin
The tasks people hand off completely are usually the ones they already found tedious or uncomfortable. HR performance write-ups. Grant eligibility screening. Client case notes.
Here's the part that should worry us. Those are also the tasks where a wrong answer causes the most damage, and where nobody is double-checking, because the work was already the kind nobody wanted to sit with. We hand off exactly the tasks where care mattered most, precisely because they were the hardest to care about.
This is why "when NOT to use AI" needs its own section in a real policy. Not folded into a general acceptable-use list. Its own space, because that's where we're tempted to stop looking.
3. Junior staff are the most exposed
The review found that people with more practice at effortful thinking were better at catching AI's mistakes. That instinct isn't innate. It's built the way any skill is built: slowly, through friction, over years of doing the hard version before the easy one existed.
New employees haven't built that judgment yet. If they onboard leaning on AI from day one, without ever doing the hard thinking the job actually requires, they may never build it at all. That's not a training gap we can patch later. It's an organization quietly losing its own capacity to catch its own errors, right when it needs more of that capacity, not less.
So what do we do about it?
David Brooks, writing in The Atlantic, described "mental marathoners," people who thrive in the AI era because they seek out cognitive complexity instead of avoiding it.
He argued we need to rethink education around that instinct, or risk a new kind of polarization: people who lean into hard thinking pulling ahead, everyone else falling behind.
Organizations face the same choice, on a shorter timeline. We can design work and culture that rewards the grind. Or we can design it to only reward speed.
In practice, that means three things: naming where AI helps and where it can't be trusted alone, protecting the tasks where a wrong answer costs the most, and treating a new employee's first years as the period where judgment gets built, not skipped.
This is the work I do with nonprofit teams at Virage: we start with your organization's values, audit how your team is already using AI, then co-create your ethical principles and usage norms together, before we ever draft the final policy.
If your organization hasn't set the guardrails for how it uses AI, and you want to avoid building a culture where cognitive surrender is acceptable, register for my next session to build your AI policy.