The tools promise to catch toxic managers before HR hears a complaint, but experts warn workplace monitoring carries real tradeoffs
A new category of workplace software promises to do what keyword filters never could. Using large language models (LLMs), these tools scan emails, chats and meeting transcripts for bullying and harassment, weighing tone, context and recurring patterns across thousands of conversations.
The software can flag bad behavior from anyone, but it carries particular weight when the offender is a manager, someone with the power to assign work, block promotions and make daily life unbearable.
Vendors such as Smarsh and Global Relay are already selling employers on the idea, marketing AI tools that promise to flag misconduct hidden in everyday workplace chatter. Toxic managers, in other words, may soon have fewer places to hide.
Whether that promise holds up is a harder question. The stakes are real. A Harris Poll survey of more than 1,300 U.S. workers in April 2026 found six in 10 currently have a toxic boss, and two-thirds have left a job because of one. Against that backdrop, three experts weighed in on whether AI detection actually helps.
The trust paradox
Robert Hurley, a professor at Fordham University's Gabelli School of Business in New York and a leading researcher on organizational trust, said the tools carry a built-in irony. Whether employees accept them depends on the very thing they're meant to protect.
"There's an ironic element to this, which is that if there's a high level of trust, that message is credible, meaning employees may believe that this monitoring is to their benefit," he said.
In a low-trust organization, the same tool lands very differently.
"Then this monitoring is problematic because people see it as Big Brother, overly intrusive, an invasion of privacy," Hurley said.
Some monitoring is necessary regardless, Hurley said. Organizations that rely purely on goodwill leave employees exposed to misconduct without consequences, which creates its own trust problem.
The bigger factor is how that monitoring gets introduced to employees in the first place. Hurley, whose research explains why trust is easy to lose and hard to rebuild, said it can't just be an FYI memo. Instead, leaders need to make the case themselves.
"Here's what we're doing, here's why we're doing it. From the C-suite, deployed down throughout the workforce, face to face," he said, adding that employees should ideally be involved in designing the system itself.
What the AI sees that humans can't
Christopher To is an assistant professor of human resource management at the Rutgers School of Management and Labor Relations in New Brunswick, New Jersey, where he teaches a new course on AI in HRM. He said the technology has two genuine advantages over human observers. It operates at scale, tracking a manager's digital behavior across years and multiple subordinates, and it sees interactions that never happen in public.
"You might be on good terms with your manager face-to-face. But, behind closed doors and in the digital space, your manager may be a very different person to one of your coworkers," he said.
But the tool has real blind spots too. Anything that happens in person, outside a transcript or an inbox, is invisible to the system. And even within its digital field of view, the tool lacks relational context.
"It's kind of like trying to judge a movie based on a few scenes. AI only sees fragments of a relationship," To said. "What looks like harsh language could be a manager setting a legitimate boundary, or joking in a high-trust relationship."
His prescription is to treat the output as a starting point rather than a verdict.
"Ideally, we use it as a first case triage but not as the final ground truth. Humans still need to be involved and review the output," he said.
To also urged HR leaders to press vendors for false positive and false negative rates, ideally verified externally, and to check performance across protected groups. The regulatory backdrop is tightening too. Illinois restricts the use of voiceprints and facial geometry under its Biometric Information Privacy Act, Maine now requires employers to notify workers about surveillance tools under a law known as LD 61, and California is weighing a bill that would require notice before deploying workplace AI tools, Assembly Bill 1898.
Where these tools fall short
Roxanne Petraeus sells compliance software for a living as co-founder and CEO of Ethena, a New York-based compliance training company. Yet her answer to whether she'd deploy AI bullying detection at her own company didn't leave much room for doubt.
"If my CPO came to me and said I want to implement AI monitoring for bullying, I would almost certainly be a no," she said. "I think that despite all of the flaws that humans have, the risk doesn't seem worth the benefit."
Part of her concern is a known weakness of LLMs. They tell people what they want to hear, a problem that also shadows the current generation of AI feedback tools.
"If I went to this tool and said, who are my three biggest bullying risks? It is absolutely going to surface three people because it knows that it wants to please the human who has asked the question," Petraeus said.
Every flag also creates work and obligation. Once an AI system reports possible bullying, the company can't simply dismiss it, so investigations pile onto the people team's desk. Petraeus draws a sharp line between using AI to investigate a complaint an employee has already raised, which she called a really appropriate use, and having AI proactively surface issues on its own, the same distinction employers face across algorithmic management systems more broadly.
Her advice for catching toxic managers is decidedly low-tech.
"It is about culture. It's about having a speak up culture where employees actually feel like they can raise issues and that when they raise them, they're going to be met seriously, with compassion, with thoughtfulness," she said.
Hurley made the same point through a different lens. AI can flag the language of bullying, but it can't tell you whether a struggling manager is coachable.
"Maybe his wife's going through cancer treatment. And this is a temporary thing," he said. "He needs somebody to put their arm around him and say, 'hey, Joe, let's take a few days off here. We care about you.' AI is not going to do that."
That may be the real takeaway for anyone weighing whether to adopt these tools. The technology can flag a problem. It still takes a person to figure out what to do about it.
