Name the tool, trace the data, identify the human choices, and connect the output to an underlying right.
‘AI made the decision’ is usually incomplete. A person chose the vendor, defined the goal, selected inputs, set a threshold, accepted an output, and decided how workers could challenge mistakes. Legal accountability lives in that chain.
Name The Actual System
A resume parser, chatbot, productivity score, scheduling optimizer, fraud flag, biometric device, and layoff model are not one technology. Each collects different data, produces a different output, and creates a different record.
Name The Actual System
Terms such as AI, algorithm, automation, and analytics can describe very different technologies. Identify the product, vendor, version, purpose, data inputs, output, threshold, user interface, decision stage, and human role. A chatbot screening tool, productivity score, scheduling optimizer, fraud flag, and layoff model create different risks. The employer may describe the tool as advisory while managers treat it as controlling. Evidence of actual use matters more than marketing language.
Trace Data To Decision
Ask what data were collected, where they came from, whether they were complete and relevant, how proxies operate, how missing data were handled, and whether the model was tested for accuracy or disparate outcomes. A worker may not have access to source code, but notices, screenshots, score changes, policy documents, vendor materials, audit logs, and human-review records can still be significant. Automated systems can scale ordinary error. They can also create common evidence across workers.
Preserve Human Accountability
Someone chooses the tool, defines the objective, sets thresholds, responds to warnings, handles appeals, and accepts the final decision. A legal analysis should not stop at 'the algorithm did it.' It should identify the employer, vendor, decision-maker, review process, and governing right. A meaningful appeal requires more than a decorative button. Record whether the worker could understand the issue, submit information, obtain human review, and receive a reasoned response.
Records That Connect Data To Decision
- Product/vendor identity, version, terms, and public technical materials.
- Notices, consent forms, policies, and screenshots.
- Inputs supplied by the worker or generated by employer systems.
- Scores, rankings, flags, reason codes, and changes over time.
- Human review, override, appeal, and audit logs.
- Outcome data and error reports appropriate to the affected decision.
Where The Analysis Can Break Down
- Not every automated tool is unlawful or inaccurate.
- Trade-secret and access limits can restrict available detail.
- Bias testing requires sound data and comparison methods.
- Privacy, biometric, disability, discrimination, wage, labor, and state laws may apply differently.
What To Do Next
- 1. Identify the tool and employment decision rather than using AI as a general label.
- 2. Preserve screenshots and notices before the interface changes.
- 3. Request a human explanation and correction route when available.
- 4. Connect the automated decision to the underlying pay, discrimination, leave, layoff, or retaliation framework.