
HR built the policies, now it needs to own the risk.
August 18, 2026

Every week for The Briefing, UNLEASH’s weekly intelligence email for senior business decision-makers, we ask our community of analysts – the true HR experts – to solve the biggest workplace challenges.
This week’s question was inspired by a recent OpEd by ‘The Office Whisper’ Dr Gleb Tsipursky on how psychological safety is quietly breaking workplace AI use.
We asked: AI policies assume people will admit their mistakes; what's the first policy change HR should make to close that gap?
Here what’s the analysts had to say.
Most AI policies carry a fatal assumption: that people will voluntarily report mistakes. They won't, not when 59% of employees already fear losing their jobs to automation.
So, what is the first policy change HR should make? It’s not a policy change but a culture change. Replace error-reporting obligations with experimentation rights.
Instead of language that says "employees must report misuse," say "employees are expected to experiment, and sharing what did not work is how we get better." That is not wordsmithing but a fundamental reframe of what the policy is for.
Pacesetter companies, the top 10% of performers in our research, already operate this way. They treat every employee as an innovator, make experimentation part of daily work, and celebrate failures as openly as wins. The best ideas come from the frontline, not from compliance teams. A warehouse worker in Japan used an agent to replace a manual inventory count, got it to 100% accuracy, and eliminated monthly weekend work.
That only happens when people feel safe to try. The goal is not a workforce that reports mistakes. It is a workforce of Superworkers who learn out loud and celebrate mistakes that make everybody smarter.
The first policy change HR should make is to replace voluntary disclosure of AI mistakes with a formal process for documenting and escalating AI errors and near misses as part of the normal workflow.
Brandon Hall Group research shows that progressive AI governance cannot rely on individual judgment alone. It calls for clearly defined responsibilities, systematic risk assessment, clear escalation procedures, audit trails, and ongoing human oversight as AI use advances, as well as builds on that principle by showing that governance must be embedded in how work is done.
Governance that operates only as policy does not function effectively at scale. Instead, AI-enabled work requires continuous governance supported by clear accountability, monitoring, quality checks, risk controls, and feedback loops. The implication for HR is to treat reporting as an operating control rather than a behavioral expectation.
Employees should have a clear way to flag incorrect, questionable, or unintended AI outcomes, with defined ownership and escalation based on the level of risk. That moves the organization from hoping people admit mistakes to systematically identifying problems, learning from them, and improving the system.
Start by asking why anyone would admit a mistake in the current environment. Surveys have found a majority of executives openly planning to sideline or lay off employees who can't or won't use AI, while a striking share of workers quietly sabotage or work around official tools.
Employees can hear both messages at once. Admitting an AI error, in that climate, is volunteering evidence against yourself. Concealment isn't a character flaw. It's a rational response to the incentives leadership built.
The first policy change is simple: separate disclosure from discipline. A no-fault reporting route for AI errors and unauthorized tool use, treated the way aviation treats near-misses, as system intelligence rather than individual failure.
When someone reports that they corrected an AI output before it shipped, or used an unsanctioned tool because the approved one couldn't do the job, that's not misconduct. That's your only accurate data about where the technology actually breaks and where the sanctioned process fails people. Every concealed mistake is a governance blind spot you paid for and can't see.
Organizations don't have an honesty problem. They have a consequences problem, and the policy has to fix the consequences first.