Developer, Deployer, or Both? Who Owes What Under AI Hiring Laws
AI hiring laws increasingly split duties between the company that builds a tool and the company that uses it. Knowing which side you are on, and whether you are on both, decides what you have to do.
Definitions
- Developer (called a provider in the EU): builds, sells or substantially modifies an AI system. HR tech vendors, assessment providers and in-house teams that build screening models.
- Deployer: uses an AI system to make or support decisions about candidates or employees. Employers and staffing agencies.
Developer obligations
- Colorado SB 26-189 (from Jan 1, 2027): give deployers documentation on intended uses, categories of training data, known limitations and how humans should review outputs; tell them about material updates; keep records for three years.
- EU AI Act (from Dec 2, 2027): provider duties for high-risk systems, including risk management, data governance, technical documentation, logging, human-oversight design, accuracy and robustness, and conformity assessment.
- Texas and Connecticut reach developers directly.
- Agency theory in the U.S.: in Mobley v. Workday, a court let discrimination claims proceed against a vendor as the employer’s agent. California’s FEHA rules likewise cover employers’ agents.
Deployer obligations
- New York City: commission the independent bias audit, publish the summary, notify candidates.
- California: anti-bias testing as evidence and four-year records under FEHA; notice, opt-out and risk assessment under the CCPA rules from 2027.
- Colorado: notice at the point of interaction, a plain-language explanation within 30 days of an adverse outcome, human review and data correction.
- Illinois and Ontario: disclose AI use to applicants and in job postings.
- EU AI Act: use the system as instructed, assign trained human oversight, monitor it, keep logs, and inform workers and candidates.
- UK and Quebec: meaningful human review of automated decisions.
Shared liability you cannot contract away
Anti-discrimination law binds everyone in the chain. Title VII and FEHA reach employers and their agents, and New Jersey’s rules apply even when the tool is vendor-supplied. Colorado goes further and voids contract terms that indemnify a developer or deployer against liability for its own violations. Vendor contracts that disclaim compliance do not move the obligation off the employer.
When you are both
- You build or fine-tune your own screening model and use it on your own candidates.
- You take a vendor model and substantially modify it, or put your own name on it. Under the EU AI Act that can make you the provider.
- You are an HR tech vendor that also uses AI to hire your own staff.
In each case you carry both sets of duties, and the evidence needs to line up: the documentation you give as a developer should match the testing you rely on as a deployer.
What good looks like
- Tag every obligation in your program as developer, deployer or both.
- Keep one evidence record that covers in-house and vendor AI.
- Ask vendors for documentation in writing, and test their tools on your own data. Our 12 vendor questions are a good start.
- Review contracts so testing, logging and audit cooperation are explicit.
Map your hiring AI against the laws that apply to you
Whether you build HR AI or deploy it, Fairly AI tests it, monitors the rules where you operate and keeps the evidence.
Schedule a discovery call