AI psychosis: Why HR cannot police its way to accountability

USCF psychiatrist Keith Sakata saw a dozen people hospitalized last year after AI use pushed them into delusion, paranoia and hallucinations. He called the phenomenon ‘AI psychosis’ and warned on social media that “we’re all vulnerable.”

A key risk factor, Sakata found, is lengthy conversation with an AI that’s overly agreeable and sounds informed – the longer the exchange, the less people reality-test what the AI says.

These clinical cases are extreme examples, but the underling pattern isn’t confined to a psychiatrist’s office – it’s already playing out at work.

SAP data found that three in four leaders trust AI’s advice over that of their colleagues and 44% would let AI override a decision they planned to make. These findings are confirmed by Glean data where 69% of leaders and workers admit not checking AI’s work, and four in ten would blame the tool if the work is wrong.

“Humans are delegating not just tasks, but judgement, to systems optimized for confidence, not truth.” They are substituting “AI’s confidence for actual human judgement and verification,” Dr Anna Tavis, Clinical Professor and Chair of the HCM Department at NYU, tells UNLEASH.

Delegating judgement to AI is a HR risk

It’s easy for leaders to fall into what Gartner HR’s Senior Director Analyst Emily Rose McRae terms “enablement illusion.” This is where AI adoption metrics create the impression that AI is delivering value; this, in turn, masks the underlying impact on the workforce.

By overlooking the potential psychological impacts of AI on workers, organizations “risk both employee wellbeing and the returns they expect on their investments.”

For Tavis, AI psychosis is a “culture problem”, not a technology one. The issue is that organizations “reward speed over verification” – this productivity pressure makes it enticing to outsource judgement to a “system optimized for confidence.”

Adler University Program Director and Associate Professor Jason Walker traces this challenge back to automation bias: the decades old tendency to accept a machine’s recommendation over our own.

What’s changed is “the machine now sounds like a colleague” – it feels fluent and confident, and it’s easy to mistake that for authority.

The outcome is a structural failure where organizations are not just adopting a tool but altering where accountability sits, according to Tavis.

Walker agrees. AI “gives a decision somewhere to hide” – “a chat window leaves nothing you can audit, so you can only fix the outcome, never the reasoning,” he says.Walker agrees. AI “gives a decision somewhere to hide” – “a chat window leaves nothing you can audit, so you can only fix the outcome, never the reasoning,” he says.

The HR fix isn’t policing or training; it’s accountability

Over-reliance on AI shows up at the top first, notes Tavis. Whether that’s with leaders who cannot “reconstruct their own reasoning,” in Walker’s words, or managers who do not push back on or fact check work from their teams.

As Gartner’s McRae notes, HR has focused too much on empowering employees to experiment with AI; this has meant “overlooking the manager’s role in driving effective use.”

HR needs to equip and train managers to be able to spot when “employees demonstrate an intense, anthropomorphized relationship with an AI tool or share delusional thinking or paranoia” – and escalate it to HR directly.

While more AI training, processes and policies are important, they alone won’t fix the over-reliance challenge.

“You cannot police your way to good judgement,” so HR acting like “workplace cops” is not the answer, notes Walker.

Whether HR likes it or not, the truth is that AI is informing more business decisions than they’re aware of. HR needs a better understanding of how employees are actually using AI.

“We’ve spent the last two years measuring whether employees are adopting AI. The next phase is measuring what AI is doing to people and to the organization,” comments Serena Huang, HR leader turned AI advisor and Founder of Data by Serena.

This does not mean creating a dedicated AI psychosis metric but measuring “the broader impact of AI alongside adoption: workload, ability to unplug, cognitive load, AI-related anxiety, sense of connection, psychological safety and whether employees feel AI is creating meaningful capacity.”

She continues: “I’m seeing more progressive HR teams use pulse surveys and focus groups for exactly these questions because the signals won’t show up in an AI usage dashboard.”

The next action is to get clear on accountability, and ensuring that “AI is a tool, never a decision maker,” states Walker. The Adler University professor recommends that organizations “name an owner on every significant decision, in writing.”

Ultimately, “accountability is what makes AI produce returns instead of confident-looking output nobody checked.”

The question the C-Suite needs to ask themselves isn’t “safety versus speed; it’s, who owns this decision and why.” “If nobody can answer, your ROI number is fiction and everyone’s agreed not to say so,” adds Walker.

Dr Tavis concludes: “The organizations that get head of this won’t be the ones who used AI the least – they’ll be the ones that made sure a human stayed accountable inside every decision created by AI.”Dr Tavis concludes: “The organizations that get head of this won’t be the ones who used AI the least – they’ll be the ones that made sure a human stayed accountable inside every decision created by AI.”