AI is becoming part of everyday workplace administration. Recruiting assistants help candidates apply. Management systems produce scores and recommendations. Recording devices turn conversations into transcripts and summaries.
Those functions are not interchangeable. A scheduling assistant may reduce coordination work, while an automated recommendation could influence someone’s employment. A recording tool may capture confidential information before anyone realizes where the data will go.
Recent California legislation gives employers another reason to examine those differences. Requirements beginning in January and July 2027 address certain AI-related workforce reductions, surveillance practices, and automated discipline and termination decisions. Healthcare employers also face practical privacy and consent questions when AI glasses enter patient-care settings.
A useful AI policy should connect the tool’s actual function to a clear owner, review process, and escalation path.
Start With the Function, Not the Product Label
“AI-enabled HR software” does not tell you enough to assess a tool.
Ask what the system actually does. Does it answer questions, organize information, rank people, flag conduct, or recommend an employment decision? Identify who sees the output and how they use it.
Gap’s recruiting assistant offers one example. The company describes Bridgette as helping candidates find jobs, apply, and schedule interviews while allowing employees to focus on conversations and decisions. That is different from allowing a system to determine who should be disciplined or terminated.
Document the business purpose before approving the product. If its use changes, review the new function rather than assuming the original approval covers it.
Give Human Review a Defined Purpose
A human approval step is not meaningful if the reviewer lacks the information or authority to disagree.
California’s SB 947 illustrates a more specific standard for covered discipline and termination decisions. Beginning July 1, 2027, employers may not rely solely on an automated decision system. When they primarily rely on its output, a human must corroborate the decision with supporting information and reject output that is unsupported, inaccurate, incomplete, or misleading.
For affected employers, that means planning the review process before the decision reaches the employee. Identify the reviewer, the supporting records, and how the review will be documented. Prepare the required notice and human-contact process separately from the general AI policy.
Outside that specific law, organizations can still use the same management questions: Who can challenge the recommendation? What do they examine? Can they stop the process?
Ask What the Tool Collects and What It Infers
The information a system generates can matter as much as the information employees enter.
California’s surveillance restrictions address AI tools that infer emotional states or collect neural data. Employers should ask vendors about features marketed as measuring engagement, stress, fatigue, mood, or attentiveness rather than reviewing only the product’s main dashboard.
Recording tools require another set of questions. Do recordings upload automatically? Are transcripts retained in personal accounts? Can the outside platform use the information for AI training? In healthcare, privacy, security, and legal teams should review the applicable HIPAA requirements and any necessary business associate agreements before authorization.
Get clear answers about the features employees can access.
Make Approval Specific
Permission to use a tool should identify the approved device, platform, account, activity, and information.
Fisher Phillips recommends this approach for AI glasses in healthcare. Approval of one documentation system should not extend automatically to personal glasses or personal AI accounts. Training should address accidental recording, automatic uploads, lost devices, and how to report a problem.
The same clarity helps managers avoid improvising rules. Explain where employees can ask questions and who can authorize a different use.
Plan for the Person Affected
An employee or candidate should have a clear route when something appears wrong.
For California decisions covered by SB 947, employers must provide a human contact and respond to requests for a meaningful, objective description of the employee data used by the system.
Healthcare facilities also need procedures for patients or visitors using recording devices, including disability-related and assistive-use requests. Staff should know when to seek help from privacy, security, or other appropriate personnel rather than trying to resolve every situation themselves.
Test One Workflow Before Expanding
Choose one existing AI use and walk through it from input to outcome.
What information enters the system?
What does the system produce?
Who checks it?
What happens when the output is wrong?
How is the issue corrected and documented?
The purpose is not to prohibit useful technology. It is to make sure the organization can explain how it is used, identify who is responsible, and intervene when the process does not work as intended.
This article provides general information, not legal advice. Consult qualified advisers about requirements applicable to your organization.