Impact
Enterprise AI hiring: why the project stalls at the compliance review
Nobody owns the answers, so the answers take eleven weeks.
October 7, 2026
Ben Chino
The short answer
Enterprise AI hiring projects are rarely rejected on the product. They stall at the compliance review, because no function owns the answers. Talent acquisition assumes Legal does, Legal assumes TA does, and IT owns only the parts it recognises. The job itself is four responsibilities and about a day a quarter.
Here is a pattern worth naming, because almost every enterprise AI hiring project hits it and almost nobody plans for it.
A talent acquisition team evaluates an AI screening or AI interview tool. The pilot goes well. The business case is solid. The project moves toward approval, and then it meets a compliance review, and it stops.
It does not get rejected. It gets postponed, pending answers. The answers take eleven weeks to assemble, by which point the budget cycle has moved and the sponsor has three other priorities.
The interesting thing is that the answers usually existed. Somebody had them. They were just never anyone's job to collect.
Why enterprise AI hiring projects stall
Enterprise buying has a stage that consumer and mid-market buying does not: a reviewer you never meet, reading a document, deciding whether signing this off is survivable. For AI hiring that reviewer sits in legal, privacy or risk, and the questions they ask are entirely predictable.
What is unpredictable is whether anyone at your company has written the answers down.
The three-way assumption
What everyone assumes
- TA assumes Legal owns it, because it is a regulatory question
- Legal assumes TA owns it, because it is TA's process and TA's vendor
- IT owns data handling and access, and assumes the employment law parts are someone else's
- Everyone is behaving sensibly
What is actually true
- No function was designed with this in mind
- It sits exactly on the boundary between three that were
- The regulations arrived looking like somebody else's problem each time
- The gap is structural, not a failure of any individual
Most compliance areas have a clear home. Data privacy sits with a named person. Financial controls sit with finance. Health and safety sits with someone whose job title says so.
AI in hiring is new enough that no function was designed around it, and it is an employment question, a technology question and a data question at the same time.
What the rules actually require
Definition
Automated employment decision tool
Any system that uses machine learning or statistical modelling to substantially assist a hiring or promotion decision. Screeners, assessments and interview analytics are all in scope.
The term comes from NYC Local Law 144 and is the phrase most US regulations have copied since.
The timeline
Four regimes, in the order they arrived
New York City, Local Law 144
Annual independent bias audit, a publicly posted summary of results, and candidate notice ten business days ahead.
California, FEHA
Bias testing, or the absence of it, becomes relevant evidence in a discrimination claim. No fine, no deadline, just a much worse position in front of a plaintiff.
Illinois, HB 3773
Notice when AI is used, no zip-code proxies for protected classes, and discriminatory effect treated as a civil rights violation.
EU AI Act, Article 50
People must be told when they are interacting with an AI system. In force now.
EU AI Act, high-risk obligations
Recruitment and selection systems. Deferred from August 2026 by the Digital Omnibus.
The one most teams have wrong
If your internal paper says you must be ready for the EU AI Act in August 2026, it is out of date. That deadline moved to December 2027. Article 50 transparency did not move, and applies today.
What ownership actually means
The whole job
Four responsibilities, about a day a quarter
It is a smaller job than it sounds, which is partly why it goes unclaimed. Nobody wants to start a function. It is about two weeks to set up and roughly a day a quarter after that.
Where to put it
There is no right answer, and the wrong answer is to leave it unassigned while agreeing it is important.
In TA, usually in a TA operations or talent systems role. The advantage is proximity: this person already knows what tools are in use. The disadvantage is they may not have standing to push back on a project the business wants.
In Legal or compliance, as a named area of responsibility rather than a new function. The advantage is authority. The disadvantage is distance from the process, which means the system list goes stale.
Split, with a single named coordinator. TA owns the inventory and the notices, Legal owns the evidence standard and reviews annually, and one named person is responsible for the whole thing being true. This is the version that survives a reorganisation.
The practical test
A company where this is owned can buy an enterprise AI hiring tool in a quarter. A company where it is not takes three.
Same product, same budget, same need. The difference is whether somebody had already written the answers down.
If you want to know where you stand, ask one question of whoever you think owns it:
How many systems in our hiring process produce a score or a ranking, and when was each one last independently audited?
A company where this is owned answers in a day. A company where it is not produces a meeting, then a working group, then a spreadsheet with gaps in it.
Key takeaways
Enterprise AI hiring stalls on ownership, not on product.
- The compliance review is predictable. Whether anyone has written the answers down is not.
- TA, Legal and IT each assume one of the others owns it. The gap is structural, not anyone's failure.
- The job is four things and costs about a day a quarter once established.
- Test it with one question: how many of our systems produce a score, and when was each last audited?
Frequently asked
Questions people ask about this
Why do enterprise AI hiring projects get delayed?
Most often at the compliance review, and not because the answer is bad. The reviewer asks predictable questions about how the score is produced, who audited it and what candidates are told, and no function owns having those answers ready. Assembling them afterwards typically takes weeks.
Who should own AI hiring compliance in an enterprise?
There is no single right answer. The three workable options are TA operations, Legal as a named area of responsibility, or a split with one named coordinator. The split tends to survive a reorganisation, because one person is responsible for the whole thing being true.
What should an enterprise have on file before buying an AI hiring tool?
Four documents: a list of every system that touches a hiring decision, the approved candidate notice wording in every language you hire in, the most recent independent bias audit per system with its date and the data it ran on, and a written answer to how you would reconstruct one candidate's score.
Does the EU AI Act still apply from August 2026?
Not for hiring. The Digital Omnibus approved in June 2026 deferred the high-risk obligations, which include recruitment and selection, to 2 December 2027. Article 50 transparency obligations took effect on 2 August 2026 and still apply.
Evaluating an AI hiring platform?
The four documents above are what your own reviewers will ask for. Ask and we will send what we give our enterprise customers to work from.
Get the working documentSources
NYC Local Law 144; California FEHA automated decision system regulations; Illinois HB 3773; EU AI Act Article 50 and the Council of the EU approval of the Digital Omnibus on AI, 29 June 2026. Last reviewed 8 October 2026.
See what Maki Agents can do for you
