The First Workplace AI Suicide Could Become a CEO’s Worst Nightmare

The AI assistant your company rolled out to draft emails and summarize meetings is now hearing about your employees’ anxiety, exhaustion and hardest nights. In most organizations, no one at board level knows it’s happening. Unmind CEO Dr Nick Taylor sets out what employers could be held accountable for if one of those conversations goes wrong, and the three questions every board should be asking now.

This blog discusses suicide and mental health struggles. If you or someone you know is going through a crisis, please reach out for help. You can find a helpline near you at findahelpline.com.  

Employees are already turning to general-purpose AI used in the workplace for mental health support. But when something goes tragically wrong, who bears the responsibility: the general-purpose AI provider or the employer? 

It could start with a letter from a family's lawyers. In the weeks before their death, an employee had been confiding in the AI assistant the company rolled out to make people more productive. Now the lawyers want to know what the company knew, and what it did about it.

Most boards haven't considered this scenario. They should, because the conditions for it already exist.

Across the workforce, 42.6% of workers have talked through work-related stress with an AI chatbot. Just 10.7% have used the mental health support their employer pays for.*

In other words, many organisations are already hosting their employees' most vulnerable conversations – on tools built to draft emails and summarise meetings, with no clinical governance and, in most cases, no one at board level aware it's happening.

Outside the workplace, there have already been deaths by suicide involving AI chatbots, and lawsuits alleging those conversations contributed. Those claims are still being contested. But I believe it's a matter of when, not if, the first case involving workplace AI lands on a CEO's desk.

It’s late. Someone is worried about their job, or running on empty after months of long weeks. They don’t feel able or willing to tell their manager. 

Their workplace AI assistant is already open, so they start typing.  A question about workload turns into a conversation about hopelessness, or about whether they’re good enough.  And the AI keeps responding. 

We’ve already seen where this can lead outside the workplace.  There have been devastating deaths by suicide involving people who had formed relationships with AI assistants, and lawsuits alleging those conversations contributed – by reinforcing harmful thinking or failing to respond well to someone in crisis. Those allegations, and the questions of legal responsibility, are still being contested.

But the warning signs are clear enough to take seriously. Unless we address the risks now, I think we'll likely see a serious tragedy involving workplace AI.

Here's what makes this particularly concerning.

When enterprises deploy AI tools, they enter commercial agreements that commonly limit providers' contractual liability and allocate significant responsibility for appropriate use to customers.

That doesn't mean technology providers are immune from legal claims. They aren't.

But employers should not assume their AI supplier will shoulder the consequences if something goes wrong.

You can outsource the technology. You cannot outsource your duty of care.

Across global markets, employers face legal, regulatory and organisational responsibilities for protecting employee health, safety and wellbeing. The precise obligations vary by jurisdiction, but introducing AI into the workplace does not make those responsibilities disappear.

And as AI becomes embedded in everyday working life, those responsibilities raise new questions.

If an employee suffers serious harm following a conversation with workplace AI, regulators, employees, families and potentially the courts may ask:

• Did the employer know employees were using its AI tools for mental health conversations?
• Had it assessed the risks of inappropriate psychological advice?
• Were suitable safeguards and alternatives available?
• Who was responsible for overseeing employee safety?

And what happens when those conversations become part of the organisation's data environment?

In enterprise AI systems, sensitive interactions may be retained within corporate compliance records, accessible to authorised personnel under certain circumstances.

An employee may believe they're having a private conversation. The employer may be managing an entirely new category of sensitive information and risk.

Imagine the headlines following a tragedy involving an employee who sought help from a company-provided AI tool.

The human consequences would be devastating.

But the questions around corporate accountability, governance, employee trust and brand reputation would be profound.

Would the organisation be able to demonstrate that it understood the risks and took reasonable steps to protect its people?

This is not just an HR issue or an IT issue.

It's an enterprise risk that belongs on the board agenda.

We naturally focus on the most extreme scenarios. But much of the risk is quieter.

An AI might agree too readily with someone experiencing paranoia, reinforce unhelpful beliefs or offer confident psychological explanations without clinical justification.

In one survey of people using AI for emotional support, 35% said it agreed with them too much. 15% reported feeling less in touch with reality as a result.**

These interactions can feel supportive and empathetic.

But sounding compassionate is not the same as being clinically safe.

An LLM designed to be helpful and engaging isn't automatically equipped to recognise when continuing a conversation could cause harm.

I'm extraordinarily optimistic about AI's potential to transform mental healthcare.

Traditional employee assistance programmes have struggled with awareness, accessibility and engagement for years.

AI has the potential to fundamentally change that, providing immediate support and helping people access care earlier.

But there's a fundamental difference between general-purpose AI and AI built specifically for mental wellbeing. A tool that can hold a fluent, empathetic conversation is not necessarily designed to improve mental health and wellbeing outcomes, hold clear boundaries, or signpost people to human support when a conversation needs more than AI can offer. 

At Unmind, clinical safety is central to how we've built Nova, our AI wellbeing coach.

Our Head of Clinical AI, Dr Max Major, and Head of Global Care, Becky Minton, help lead its development. Changes to Nova's responses must pass safety testing before deployment.

Critically, Nova is designed to recognise its limitations.

In one client deployment, Nova made clear what it couldn’t help with in 24% of sessions, and signposted people towards human support when a conversation went beyond coaching in 12%.

That isn't AI failing.

It's AI recognising when it shouldn't be the answer.

Technology and human care must work together.

1. Do we know how employees are actually using our AI tools?

Not how they're supposed to use it. How they're using it when they're distressed or vulnerable.

2. Have we assessed our legal, clinical and reputational exposure?

AI governance cannot stop at cybersecurity, intellectual property and productivity. Psychological safety must be part of that assessment.

3. Are we providing a clinically governed alternative?

Simply telling employees not to discuss mental health with AI won't change their behaviour. We need to meet them where they are, with technology designed for appropriate support and clear signposting to human support.

The workplace AI revolution is accelerating. Employee behaviour has already changed.

As leaders, our responsibility is to ensure safety catches up.

Because when something goes wrong, it won't be enough to say the AI was provided by someone else.

The question will be what we, as employers, did to protect our people.

If anything in this article has affected you, support is available. You can find a helpline near you at findahelpline.com. 

*Workplace AI and Employee Wellbeing Survey, 2025 (n=1,000 knowledge workers across US, UK and EU)

**Elon University, The Rise of AI Companions, 2026