Dario Amodei has spent much of the past year making the case that the industry should not sprint blindly toward more capable AI. The Anthropic chief executive has called for deliberately pacing the development of frontier models so that safeguards have time to catch up, proposing independent evaluators, stronger alignment practices, and deeper coordination between companies and governments.
Writing in American Thinker, Nancy E. Parsons accepts that those concerns are justified. But she argues the debate is missing a source of risk that is hiding in plain sight: the psychological makeup and motivational drivers of the teams building, governing, and deploying AI in the first place.
Her question is blunt. If the industry is willing to spend billions evaluating what is inside the models, who is evaluating what is inside the humans making the decisions?
The problem isn’t one person. It’s the room.
Parsons, who says she has spent more than three decades studying personality strengths, risk tolerance, motivation, and leadership team composition, is careful to locate the danger in the collective rather than the individual. The greater threat, she writes, resides in homogeneous teams.
When the same behavioral predispositions are concentrated across a leadership or development group, the natural counterweights disappear. Who slows the decision down? Who questions the assumption everyone else has already accepted? Who thinks about the unintended consequences, or insists on a boundary when the rest of the room is energized by breaking one?
The urgency, in her telling, comes from timing. Humans remain substantially in control of AI’s trajectory, and capabilities are advancing quickly. Prevention has to happen while human decisions can still meaningfully shape where the technology goes. The guardrails worth examining right now, she suggests, are the human ones.
Brilliance isn’t the same as judgment
There is a tension at the heart of Parsons’ argument that she spells out plainly: the characteristics that make people exceptional at pushing boundaries may make them the wrong people to set those boundaries. Technical brilliance should qualify someone to build AI. It should not, on its own, qualify that person to decide how much risk society ought to accept from it.
That is not a case for replacing daring innovators with cautious ones. Parsons frames it as a case for balance. When nearly everyone at the table is highly inquisitive, adventurous, comfortable with significant risk, willing to break rules, and energized by power and competition, a single cautious voice is not a counterweight. That person, she writes, is likely to be ignored or shut down by the prevailing culture.
What’s needed instead are multiple credible counterbalances with the authority to actually influence decisions — and a leadership team aligned around a mission that spells out not just what it hopes to accomplish but what risks it is unwilling to take. Getting there requires assessing the collective human risk of leadership and R&D teams and deliberately building counterbalancing capabilities where they are missing.
Parsons is skeptical that standard executive evaluation can get the job done. Résumés, interviews, performance histories, and background checks all say something about experience and observable behavior, but she argues they offer limited insight into inherent behavioral tendencies, talent gaps, or the imbalances that put an organization at risk. Scientifically valid measures, applied individually and collectively, are what she wants to see — before the risks become consequences.
An Enron comparison, and a Silicon Valley case study
This is not a new argument for Parsons. More than two decades ago, she made a version of it about corporate risk. In an article for Risk Management Magazine, she compared assessment data from former Enron executives with data from felons incarcerated in a maximum-security prison and found the profiles remarkably similar. Her conclusion at the time was that companies were diligently evaluating financial, operational, and market risks while largely ignoring the measurable behavioral risks sitting inside the executive suite.
The AI industry, she suggests, is repeating that pattern at a much higher stakes level.
She offers one recent example. Parsons says she facilitated a development session with the C-suite of a Silicon Valley AI company, and the assessment data revealed an extraordinarily homogeneous team: brilliant innovators and risk-takers, with very little counterbalancing operational capability. The executives saw it immediately, she writes. The same strengths that fueled their innovation were also feeding execution problems and putting the business in jeopardy.
Their response is worth noting, because it cuts against the idea that this kind of assessment inevitably leads to tamping down the boldest people in the room. They didn’t constrain the innovators. They brought more operational, practical, results-oriented talent up from the level below into the leadership conversation. Balance, in other words, rather than less brilliance.
Parsons maintains that the tools to do this already exist — to assess AI executive and R&D teams, to identify behavioral strengths and risks, and to determine whether a given team has the counterbalances required for sound decision-making. The open question, as she puts it, is whether anyone will actually use them.
Her closing framing is that AI presents an old human problem at a new scale and speed. The principles aren’t novel; the stakes are. If the goal is for AI to serve humanity well, she writes, the industry must pay as much attention to the judgment, balance, and boundaries of the people leading the technology as it does to the technology itself.
Source: www.americanthinker.com — https://www.americanthinker.com/blog/2026/09/the-biggest-risk-from-ai/