Even agentic HR developers are split on using AI to report workplace misconduct

Sponsored by

What’s new with HR tech? A lot, actually. To better understand AI readiness in HR, HR Brew explores the state of the industry, what’s being prioritized with AI, and what HR teams are doing to stay prepped. Read it here.

As AI permeates all aspects of corporate people ops, HR leaders are grappling with the possibility of delegating to agents even the most harrowingly human aspects of their jobs—namely, responding to workplace misconduct, harassment, and discrimination reports from employees.

But this is not purely a human-versus-machine issue; it’s fraught for the software companies that engineer HR agents, too, as they test new boundaries with agents that can direct employees to human HR representatives in response to misconduct complaints, or “empathize” with them and gather information about the reported incident.

Morning Brew spoke with five HR-specific autonomous agent providers to elucidate how AI is being deployed at different stages in the misconduct reporting and investigation process. Proponents of the tech argue that reporting incidents to AI can ensure the misconduct complaint is handled in a uniform way and can further encourage reporting. On the flip side, one workplace AI company we spoke with said it wouldn’t use agents for such a thing, because when it comes to workplace misconduct, people want to speak with a human.

And beyond the human element, there’s a legal one: Employers that deploy AI for misconduct reporting can encounter significant liability risks and potentially undermine the severity of a complaint by asking employees to report painful incidents to a non-human entity, attorneys told us.

Agentic colleagues

Beyond answering questions about company policy and autonomously handling admin work, some HR agents can respond to reports of workplace misconduct in more nuanced ways. Harper, an agent created by HR platform Wisq, is “a representative of HR” that can escalate initial complaints to humans, per the company’s head of product, Clayton Holz. He told Morning Brew that workplace misconduct reporting is “actually a good use case for AI” because it can be more uniformly trained than humans.

“If you think about having maybe 50 different folks in the HR department who could be doing these interviews, they’re on site, there’s turnover in this role, they don’t have a lot of experience doing it,” Holz said. “The consistency of AI, as well as the fact that everything’s well documented, it could be net beneficial.”

Holz said he thinks AI could create plans for evaluating complaints and even executing parts of the investigation. (While these aren’t functionalities Harper currently has, Holz said some clients are asking for them.) Giving agents more responsibility in this process could trigger pushback and concerns about AI bias, but Harper has gone through bias testing, Holz said, and even escalates complaints to humans more often than might be necessary.

“We have a dedicated [AI] model that is trained on tens of thousands of examples to specifically pluck [misconduct reports] out, and then say, ‘I need to pass this on to my colleague,’” Holz said, adding that Harper has a “bias toward false positives and not false negatives. No AI model is perfect, and anyone who tells you otherwise is lying to you.”

Making sure that agents are role-specific helps them stay on track, too. Lattice Chief Strategy and Marketing Officer Barbra Gago told Morning Brew in a statement that though the company’s HR-specific AI agent isn’t currently “detecting or autonomously escalating employee misconduct,” “companies benefit from using purpose-built enterprise AI” rather than a “general-purpose assistant.”

“The challenge isn’t simply giving an AI access to HR information,” she said, “it’s building the governance, permissions, context, auditability, and human oversight required to use AI responsibly in sensitive workplace situations.”

A systematic actor

Sophie, an agent from customer experience software Konnect, takes on more of the reporting process than Wisq’s Harper. When Sophie receives a misconduct complaint, it first asks the employee if they’d like to report the incident confidentially or anonymously. Then, Sophie gathers information, like what happened, who was involved, and when it occurred, much like a whistleblower hotline.

If the employee decides to report anonymously, “there’s no way that the tool or the [employer] can trace it back to whoever’s reporting that issue,” Konnect Founder and CEO Jamie Viramontes said. The agent gives the employee a tracking number so they can check on the progress of their complaint and “deletes the data they reported from the server.” HR representatives can then communicate with the employee via the tracking number, which also allows them to anonymously respond to follow-up questions.

Once a workplace misconduct report is made, the human HR team is notified of a “high” alert in Sophie’s system.

“If you see a high-priority issue come through, you should get to it immediately,” Viramontes said he tells clients. “They get an email. They could get a text. They could get some alert that says this issue is happening live, and you need to get on it right now.”

If an employee decides they no longer want to report the issue, the agent will still submit the complaint with the information it gathered, and if at any point the employee wants to stop answering the agent’s questions, they can click a “send report now” button. And, if employers choose to include it, employees can also tell Sophie they want to speak with a human and be immediately connected with an HR representative.

“Anything that comes through the employee relations module, all of that gets to a human for it to be investigated and followed up on,” Viramontes said. “It doesn’t get left just sitting.”

Legal liability

Should an agent fail to escalate a report to a human, though, the company using an HR agent could be held responsible.

Dave Walton, a partner at Fisher Phillips and the chair of the law firm’s artificial intelligence team, told Morning Brew that once a misconduct report is made, if for some reason the agent doesn’t communicate it to a human, the company would still be considered “on notice.” Agents also need to quickly deliver complaints to the right human to ensure swift remedial action.

“Once that misconduct is reported to the agent, I think the courts are going to assume that’s notice to the whole company,” Walton said. “The company can’t say, ‘Well, the agent didn’t tell me this. I didn’t know.’ That’s going to fall on the company [because] you didn’t set your agent up the right way.”

And in the event of a lawsuit, agents dramatically increase the amount of electronically stored information (ESI), all of which is discoverable evidence in a case and can be “hard to control,” Walton noted. An agent’s ESI also includes its skills, or the text file that instructs the agent on how to respond to different commands or prompts, Walton said.

“It becomes an evidentiary issue” for the company using the agent, he said.

Adit Jain, the CEO and co-founder of Leena AI, told Morning Brew he discusses these legal realities with his customers. The company’s HR agent, Harrison, has three escalating configurations for employee complaints, including having the agent listen to the complaint and—if the employee doesn’t want it escalated to a human or doesn’t reach out to a human themselves—“not do anything with it.”

“Our customers make those choices themselves after talking to their legal teams,” Jain said. “I’m presuming if this is a big legal risk, they would turn [automatic reporting] on.”

Distributing the risks that come with using agents for sensitive conversations is also a big reason to buy enterprise AI software rather than building it in-house, Lydia Wu, Gartner senior director and industry analyst on AI in HR, told Morning Brew.

Vendors and companies can work together to ensure that agents stay on track, rather than companies taking on “100% accountability” should anything go wrong. Building agents in-house, she said, “creates a layer of additional corporate risk that not many HR teams are willing to tolerate.”

Even when companies buy enterprise software, though, Wu said the return on investment of deploying AI for misconduct reporting could end up being “marginal.” With “messy, human-involved use cases,” AI shouldn’t be involved—“at least not now.”

“AI is not peanut butter,” she said. “You can’t just spread it across your whole [HR] function and hope that it’s gonna work.”

The human cost

Some HR leaders agree with Wu’s analysis and feel that workplace misconduct conversations are too sensitive to hand over to AI—even in part. Laura Gonzalez, the chief of staff and head of people at AI video platform Synthesia, deploys AI-generated agents in the form of avatars for communications with employees. But, she told Morning Brew, she draws the line at using it for submitting misconduct complaints.

“If someone has a real concern about there being misconduct, they will want to talk to someone in the HR team. They will not want to report it to an avatar,” she said. “We invested a lot in enterprise readiness, permissions, consents—everything. But I think there’s just certain conversations that you would really want to talk to a human about.”

But she’s a booster for deploying AI in other HR realms: It frees up space for the company’s HR team to “have more dedicated bandwidth on a human level for the things that really need to happen on a human level.”

And if conversations surrounding workplace misconduct aren’t human-to-human, employers risk exacerbating the issue, public interest lawyer Peter Romer-Friedman told Morning Brew. He’s the founder and managing partner of Romer-Friedman Berman Law, and he said some of his clients found reporting misconduct “very disempowering” because they feel that many HR departments disincentivize them from doing so.

“When you’re talking about someone where there’s a real severe problem going on, it’s so demeaning to have to give that information to an AI agent,” Romer-Friedman said. “A lot of people might come away from that experience, and the next phone call they make is to a lawyer like me.”

There’s also a lot of value in reporting difficult experiences to a human being, he said—in part because doing so can galvanize the company into taking remedial action quicker.

“If you’re complaining about something very personal, like being inappropriately touched or assaulted, you shouldn’t have to speak to an AI agent. There should be a human being on the other end of that conversation to empathize and sympathize and ensure that there’s immediate action taken—the person who picks up the phone after that call and says, ‘We’ve got to do something right now,’” Romer-Friedman said. “Would an AI agent do that? Or is it just going to put your information in a queue for some person to read?”