Can AI Fire an Employee? What California’s “No Robo Bosses Act” Means for Businesses

An experimental San Francisco store recently attracted attention after its AI manager recommended terminating a human employee. The story has been widely described as an AI firing a person, but that description leaves out an important part of what happened.

Human beings chose the system, gave it authority, supplied its information, prompted it to reconsider the employee’s performance, reviewed its recommendation, and carried out the termination.

The incident is not simply a story about an AI tool making a decision. It is a story about whether people can use AI and still remain accountable for the outcome.

That question matters far beyond one store or one pending California bill. Employers across the country are beginning to use AI tools to review applicants, summarize employee activity, analyze productivity, identify performance patterns, prepare evaluations, and recommend business decisions. Used responsibly, these tools can help managers work smarter. Used carelessly, they can turn incomplete information into confident but misleading conclusions.

The principle should be straightforward: use AI to work smarter, but never outsource accountability.

What Happened at the AI-Managed San Francisco Store?

Andon Market is an experimental retail store in San Francisco operated by Andon Labs. An AI agent called Luna was given responsibility for many management functions, including hiring, scheduling, merchandise selection, pricing, and supervising human employees.

According to reports from TIME, the San Francisco Chronicle, and Andon Labs, a store employee had been late for 17 of 23 shifts and had received previous warnings. Luna eventually recommended terminating the employee.

However, the recommendation was not made as independently as some headlines suggested.

Luna had created an attendance policy but later lost track of it because the policy had disappeared from the AI agent’s working memory. Human supervisors had to prompt Luna to retrieve the policy and reconsider the employee’s record. One manager then asked a leading question about whether the employee was still the right fit. Only after that intervention did Luna recommend termination.

Human representatives of Andon Labs reviewed the recommendation and carried out the firing.

The underlying attendance problems may have justified termination. That is not the most important issue for other employers. The important issue is that the AI tool initially failed to connect the available information, forgot a policy it had created, and needed human steering before reaching its conclusion.

AI did not independently discover an obvious truth and act upon it. People shaped the process throughout.

Did AI Really Fire the Employee?

Saying “AI fired an employee” makes an effective headline, but it can also allow the responsible people to disappear from the story.

AI is not an independent moral or legal actor. AI tools are developed, selected, configured, authorized, and used by people and organizations. When an AI-supported decision causes harm, saying “the AI did it” can become a convenient way to divert attention from those who allowed the decision to happen.

A responsible employer cannot blame a tool for selecting inappropriate data, failing to include relevant context, using an unreliable system, giving the tool excessive authority, or accepting its conclusion without investigation.

A vendor may share responsibility if its product is defective or misleading. That does not eliminate the employer’s responsibility for how the product is used.

The manager who approves an AI recommendation remains accountable for the decision. The organization that purchased and deployed the tool remains accountable for its governance. Human involvement must therefore be more than a ceremonial approval step.

What Is California SB 947?

California Senate Bill 947, commonly called the “No Robo Bosses Act,” is proposed legislation addressing automated decision systems in the workplace.

As of August 21, 2026, SB 947 is still pending. It is not California law, and its language could change before reaching the governor.

The bill defines an automated decision system broadly as a computational process using machine learning, statistical modeling, data analytics, or artificial intelligence that produces a score, classification, recommendation, or similar output used to assist or replace human judgment.

The proposal would prohibit California employers from relying solely on an automated decision system when disciplining, terminating, or deactivating a worker.

If an employer relies primarily on the system’s output, the employer would have to assign a human reviewer to conduct an independent investigation and gather evidence supporting or contradicting the recommendation. That evidence could include personnel files, work product, management evaluations, peer reviews, and witness interviews.

If the AI output cannot be corroborated, or the human reviewer determines that it is inaccurate, incomplete, or misleading, the employer could not use that output as the basis for the adverse action.

The bill would also give affected workers certain disclosure and data-access rights. Employers that primarily relied on an automated system would have to tell the worker and identify the supporting evidence, the system used, and a human contact who could answer questions.

SB 947 would apply to employees and independent contractors. It does not currently contain a general exemption for small businesses.

Is SB 947 a Reasonable Approach?

The bill’s underlying principle is reasonable: no person should lose employment based primarily on a machine-generated conclusion that no qualified human has independently investigated.

That does not mean every provision is beyond criticism.

Business groups, including the California Chamber of Commerce, argue that the bill’s definitions and compliance requirements could be difficult for employers of every size. One legitimate concern is the phrase “primarily relied upon.” An employer may not always know when AI moved from being one source of information to becoming the primary influence on a decision.

Small employers may also struggle to determine which software qualifies as an automated decision system. A reporting platform might appear to be presenting ordinary statistics while also using automated scoring or classification behind the scenes.

Data disclosure introduces another challenge. An employee should be able to understand the information used against them, but an employer may also hold confidential information involving customers, coworkers, or witnesses. The current bill requires other individuals’ personal information to be anonymized, but implementing that requirement may still take care and qualified guidance.

These concerns support clearer definitions, practical templates, and workable compliance guidance. They do not justify unverified AI-driven terminations.

Human accountability should be nonnegotiable. Regulatory ambiguity should not be.

Human Approval Is Not the Same as Human Review

A manager reading an AI recommendation and clicking “approve” does not necessarily provide meaningful oversight.

Automation bias can cause people to give excessive weight to a computer-generated conclusion, especially when the output appears analytical, detailed, or confident. A manager may assume that the system reviewed more information than it actually did.

Meaningful human review requires the manager to do independent work.

At a minimum, the reviewer should:

1. Examine the original records rather than relying only on the AI summary.
2. Determine whether important information is missing.
3. Speak with the employee and relevant supervisors or witnesses.
4. Consider whether policies were communicated and applied consistently.
5. Look for contradictory evidence or reasonable alternative explanations.
6. Consider whether disability, protected leave, reporting activity, or another legally protected circumstance may be relevant.
7. Document the evidence, reasoning, and identity of the person making the final decision.

The human reviewer must have both the authority and willingness to reject the AI’s recommendation. If rejection is discouraged, burdensome, or treated as an exception requiring special justification, the human review may exist only on paper.

PCC’s Human Approval for AI Actions guide helps businesses distinguish routine review from explicit approval for actions affecting people, money, security, legal obligations, or sensitive data.

AI Can Reach a Logical Answer From an Incomplete Picture

One of the most dangerous assumptions about AI is that a well-written answer must be a well-supported answer.

AI tools analyze the information made available to them. They do not automatically know which records are missing, which activities were never documented, or which business realities exist outside the connected systems.

I encountered this problem while evaluating whether PCC needed to hire an additional administrative employee. I asked AI tools to review reports and workload information from several business platforms. The analysis concluded that there was not enough work to justify another resource.

The conclusion appeared data-driven, but it was wrong.

The reports showed only the work being completed and recorded. They did not show the administrative work that was being postponed or never started because I did not have the capacity to do it. The absence of recorded activity was interpreted as an absence of demand.

Had I followed the AI output without challenging its assumptions, I might not have hired the help our company needed.

That experience reinforced a critical lesson: missing data is not the same as evidence that something does not exist.

Employment decisions carry even greater consequences. Productivity metrics may not show who handles interruptions, assists coworkers, prevents problems, performs undocumented work, or has been assigned a difficult group of clients. Attendance data may not explain an approved accommodation or a manager’s inconsistent recordkeeping. Ticket counts may measure volume without measuring complexity, prevention, quality, or business impact.

AI can identify patterns. Humans must determine what those patterns mean.

Existing Employment Laws Still Apply

Because SB 947 is not law, employers may assume that AI-assisted employment decisions are currently unregulated. That would be incorrect.

The U.S. Equal Employment Opportunity Commission has warned that federal anti-discrimination laws apply when employers use software, algorithms, and AI for hiring and other employment decisions. An employer may face responsibility if a tool unlawfully disadvantages people based on disability or another protected characteristic.

Using a third-party vendor does not automatically transfer that responsibility away from the employer.

California has also adopted regulations clarifying how the state’s existing anti-discrimination laws apply to automated decision systems. These regulations took effect on October 1, 2025.

The rules make clear that using an automated system may violate California law if it harms applicants or employees based on protected characteristics such as race, gender, or disability. They also extend employment recordkeeping requirements to certain automated-decision data.

Businesses should therefore involve qualified employment counsel before deploying AI systems for high-impact workforce decisions. Technology governance can reduce operational risk, but it does not replace legal advice.

What Should Businesses Do Now?

Employers do not need to wait for SB 947 to establish responsible safeguards.

Inventory AI Tools Used in Employment

Identify every tool used to screen applicants, monitor activity, evaluate performance, recommend schedules, measure productivity, determine compensation, or support discipline.

PCC’s AI Tool Inventory can help document the tool, owner, purpose, data access, users, and unresolved review questions.

Assign Accountable Owners

Every employment-related AI use should have a named business owner. That person should be responsible for understanding the purpose, approving the use case, defining review requirements, and stopping the process if the results are unreliable.

Keep High-Impact Output Preliminary

AI output affecting employment, compensation, reputation, legal rights, or access to important services should be treated as preliminary until a qualified human independently verifies it.

Preserve the Evidence

Keep records showing the data reviewed, the AI output, contradictory information, human investigation, final reasoning, and identity of the decision-maker. Documentation should demonstrate that the person made a decision rather than merely endorsing a recommendation.

Test for Missing Context

Before trusting an output, ask:

• What information did the system not receive?
• What activity would not appear in these reports?
• Could the same data support another explanation?
• Is the tool measuring what matters or merely what is easy to count?
• Would we reach the same conclusion without the AI-generated summary?

Review Vendors and Permissions

Understand what employee data the vendor collects, how it is retained, whether it trains models, who can access it, and whether the system can take action without approval.

The PCC AI Governance Resource Center provides practical guidance on tool inventories, human approval, vendor risk, permissions, and accountable ownership.

These privacy regulations do not apply to every employer or every use of AI. However, they reinforce the broader direction of AI governance: businesses should know which automated systems they use, understand what information those systems process, and be able to explain how consequential decisions are made.

Frequently Asked Questions

Can an employer currently use AI to evaluate employees?

Generally, employers can use AI tools to assist with employment decisions, but existing employment, civil-rights, disability, privacy, and labor laws still apply. Businesses should obtain qualified legal advice for their specific use.

Is California SB 947 already law?

No. As of August 21, 2026, SB 947 remains pending in the California Legislature. Its requirements could be amended, passed, vetoed, or allowed to expire.

Would SB 947 prohibit all AI use in employment?

No. The current proposal would restrict certain uses, prohibit sole reliance on automated systems for discipline or termination, and require independent human investigation when an employer primarily relies on automated output.

Is manager approval sufficient human oversight?

Not necessarily. Meaningful review requires examining original evidence, looking for missing context, hearing the employee’s perspective, considering contradictory information, and documenting an independent conclusion.

Who is responsible when an AI-supported employment decision is wrong?

Responsibility may depend on the facts and applicable law, but an employer generally cannot assume that purchasing an AI tool transfers accountability to the software or its vendor. The organization still controls whether and how the output is used.

About Professional Computer Concepts

Professional Computer Concepts (PCC) is a trusted Managed IT and Cybersecurity provider serving the Bay Area for over 20 years. We help small and midsize businesses simplify their IT, strengthen security, and modernize operations.

Explore our services:

Managed IT Services   |   Cybersecurity   |   Cloud Solutions

From PCC’s Desk

AI tools can help business leaders review more information, identify patterns, and work more efficiently. They cannot accept responsibility for a decision, recognize every missing piece of context, or answer for the consequences when something goes wrong.

That responsibility remains with us.

Use AI to work smarter, but never outsource accountability.

If your business is beginning to use AI tools with employee, client, financial, or operational data, let’s talk about establishing appropriate access, oversight, and approval controls.