# Does a tool update need a new audit: Local Law 144 says 1 year, not each patch

Sarah Johnson · November 16, 2026

> NYC Local Law 144 requires an independent bias audit at least once per year; individual tool patches do not automatically require a new audit.

| Takeaway | Detail |
| --- | --- |
| NYC Local Law 144 mandates a bias audit at least once per year. | Employers must ensure an independent bias audit of each automated employment decision tool has been conducted within the prior year. |
| Individual patches do not automatically require a new independent audit. | The annual audit cycle set by Local Law 144 does not mandate commissioning a separate audit after every tool update. |
| Bias audits must be performed by an independent auditor. | Employers using an automated employment decision tool must commission an independent audit rather than conducting a self-assessment. |
| Tool updates that affect audit validity may require a new audit before use. | Employers should verify whether an update changes the tool in a way that makes the existing annual bias audit no longer representative. |

NYC Local Law 144 requires employers to commission an independent bias audit of an automated employment decision tool at least once per year, not after each patch.

That answer changes if a tool update alters the tool in a way that undermines the validity of the existing annual audit.

![Sunlit municipal corridor with pale stone floors glass](https://static.mm-ais.com/article-images-ai/does-a-tool-update-need-a-new-audit-loca-ai-bfcf24e2.jpg)
Sunlit municipal corridor with pale stone floors glass

## How It Works

Check the current official text for when the bias-audit duty applies and what compliance date governs. After a software change, the live question is whether the updated system is still the same tool that was audited, or whether the change altered the model, inputs, scoring logic, or decision output enough that the existing audit no longer matches what you are using. Verify the current build, the current use case, and the current audit scope before you commit to “no new audit needed.”

Before relying on these terms, verify their current official definitions. Check whether the system is an “automated employment decision tool” for the way it is actually used, whether it substantially assists or replaces discretionary decision-making, and what qualifies as an “independent bias audit.” Confirm that the audit record uses the tool’s actual decision data and selection-rate calculations, rather than vendor branding or a summary claim.

Independence is also part of the mechanism. The audit must be conducted by an independent auditor, and the audit record should let you verify who performed it, what data was used, what groups were examined, and what selection and impact-rate calculations were reported. Keep the report, the auditor’s qualifications, the tool version, and the internal decision record together. If those items do not line up, you do not yet have a clean basis for saying the post-update tool is covered by the earlier audit.

Use a simple rule: verify the live configuration first, then compare it to the audit’s stated scope. If the update changes only presentation or administrative mechanics, the existing audit may still align with the tool that was reviewed. If the update changes how candidates are evaluated or how decisions are made, treat the earlier report as a document to re-check against the new system rather than as automatic coverage. That is the verify-before-you-commit step: match the tool you actually run to the tool the audit actually assessed.

![Quiet urban street dawn after light rain brick](https://static.mm-ais.com/article-images-ai/does-a-tool-update-need-a-new-audit-loca-ai-1bdee404.jpg)
Quiet urban street dawn after light rain brick

## Key Factors to Consider

When you verify before you commit to a fresh bias audit after an update, start with the three decision criteria that matter most: whether the change alters the tool’s scoring or selection logic, whether the change affects the population or job category it is used for, and whether your current audit record still matches the live version you are using. Treat these as a single checklist, not three separate debates. If any one of them is unresolved, hold the commitment decision until you can compare the live, complete option against the version your existing documentation actually covers.

First, check the substance of the update. A cosmetic interface revision, a hosting migration, or a documentation edit may leave the decision process intact, while a change to model weights, ranking rules, cutoff behavior, or the features fed into the tool can change how candidates are evaluated. The practical rule is like-for-like: compare the old and new versions at the level of inputs, outputs, and decision influence. If the updated tool still substantially assists or replaces the same employment decision in the same way, your existing audit may remain relevant; if the decision logic has shifted, the case for a new independent audit gets stronger.

Second, check the scope of use. An audit is not a general-purpose badge that follows a tool anywhere it is deployed. Verify the job categories, applicant pools, locations, and decision stages covered by the audit you are relying on. If the updated tool is now used for a different role family, a broader candidate group, or a more consequential stage of the hiring process, the old evidence may no longer be a like-for-like match. This is where the reader rule matters most: confirm the live, complete option before committing, including the audit’s stated scope and the employer or employment agency’s actual use of the tool.

Third, check independence and record quality. Check the current official text for the independence requirement and the materials that must be available. Before you commission or reuse anything, verify who performed the work, what version was tested, what data period was used, and whether the result is complete rather than a summary or marketing claim. A clean-looking certificate is not the same as a complete audit record. Ask for the full package and compare its identifiers, dates, and stated limitations against the tool you intend to use.

The numbers that matter here are mostly verification numbers, not guessed costs. Capture the audit date, the tool version tested, the data period covered, the number of job categories or decision groups in scope, and the selection or impact rates reported for each relevant group. Recompute any rate you rely on from the underlying counts before treating it as comparable. If the source gives a percentage without counts, request the counts. If it gives counts without a clear data window, request the window. Do not fill gaps with assumptions.

Use current, sourced figures only when they are directly relevant to the audit record. For audit decisions, rely on the live tool record, the complete audit document, and a like-for-like comparison of what was tested versus what you are about to use. If outside workforce data is offered, verify its date, source, and relevance before using it in the comparison.

![Key Factors to Consider — Does a tool update need a](https://static.mm-ais.com/article-images-pixabay/does-a-tool-update-need-a-new-audit-loca-752dceb1.jpg)

## Common Mistakes

Two mistakes show up again and again when teams verify whether a software change calls for a fresh independent bias audit. Both are avoidable if you slow down and compare the live, complete option in front of you instead of reacting to the update notice.

The first pitfall is treating any vendor release note as a self-executing audit trigger. A concrete example: a vendor emails that it changed the wording of a candidate-facing explanation, added a new dashboard chart, and refreshed its help text. The update is real, but the change is cosmetic if the scoring model, cutoff behavior, and candidate population handled by the tool stayed the same. Committing to a new audit on that basis burns time and budget without answering a compliance question. The check is to request the current release documentation and confirm what actually changed in the decision path, not just what changed in the interface.

The second pitfall is the mirror image: assuming an audit from an earlier version still covers the tool after a substantive change has been quietly folded in. That can happen when a vendor retrains a model, changes a ranking rule, or expands the tool to a different role family while the employer keeps using the same product name. The label stays familiar, but the thing being used is not the same thing that was audited. Verify the version identifier, the audit scope, and the job categories covered before relying on the older report.

Also verify the independence and completeness of the report you are relying on. A summary slide or marketing badge is not the same as the full audit record. Ask for the complete document, confirm who performed it, and check that it addresses the tool version you are actually using. Use the same like-for-like comparison described above, and do not accept a headline when the underlying scope is missing.

Finally, keep the decision tied to use, not to calendar habit or product branding. An update matters when it changes how the tool substantially assists or replaces employment decisions for the population and roles you cover. It does not matter merely because the vendor shipped something. Document the comparison, preserve the release materials you relied on, and make the call from the live configuration rather than from the announcement.

![Common Mistakes — Does a tool update need a](https://static.mm-ais.com/article-images-pixabay/does-a-tool-update-need-a-new-audit-loca-f4f60e91.jpg)

## Insider Tactics

Insider tactics here are about sequencing, not clever wording. The non-obvious move is to treat the vendor’s release notes as evidence to be preserved, not as the decision itself. Before anyone signs a new scope of work, capture the exact version in use, the exact version proposed, the stated functional changes, and the vendor’s own characterization of whether scoring, selection, or eligibility behavior changed. That packet becomes your verification file. If the vendor later says the update was “administrative” or “cosmetic,” you can test that statement against the contemporaneous record instead of relying on a support ticket memory.

A second tactic is to separate “new audit” from “new evidence.” Sometimes the practical answer is to ask the independent auditor to perform a targeted re-review tied to the changed component, rather than commissioning a full replacement engagement from scratch. The point is not to assume that option is legally sufficient in every case; the point is to put the question to the auditor in writing with the change packet attached. A named independent reviewer can tell you whether the prior work can be relied on for unchanged parts and where a fresh selection-rate or impact-rate analysis is needed. That written response is useful even if you ultimately buy a full audit, because it shows the decision was made with expert input.

The timing tip is to start the verification clock when the update is proposed, not when deployment is scheduled. Procurement, legal review, and auditor availability can take longer than the engineering sprint. Build a simple internal checkpoint: product owner confirms what changed, HR confirms which jobs and applicant pools are affected, privacy or records staff confirms what data fields moved, and procurement confirms whether the existing audit contract has language for follow-on work. If those answers are not ready, hold the release decision open. This is especially useful where the vendor bundles a model refresh with interface changes, because the bundle can hide a substantive scoring change inside an ordinary-looking upgrade.

Another timing move is to align the verification with your ordinary change-control meeting rather than creating a one-off compliance scramble. Put the AEDT update on the same agenda as security review, accessibility review, and data-retention review. That gives you a single place to ask whether the update changes outcomes, not just whether it compiles and ships. It also creates a clean paper trail showing that the employer considered the bias-audit question before commitment, which is the practical value of verify-before-you-commit.

Use a like-for-like comparison when you price or scope the response. If one option is a full independent audit and another is a targeted update to an existing audit, require both to address the same tool version, the same job categories, the same decision rates, and the same documentation standard. Do not compare a broad proposal from one provider with a narrow memo from another and call the cheaper one efficient. The reader rule is simple: verify the live, complete option before committing, then compare totals and terms on the same basis.

One final insider check is to name the person who owns the final call. In practice, the safest workflow is not “engineering approves, legal later reviews.” It is “engineering documents, HR validates use, independent auditor advises, and a designated compliance owner signs off.” That last signature should be dated before the updated tool is used for employment decisions. If the sign-off cannot be produced on time, the update is not ready for commitment, regardless of how routine the vendor says it is.

![Insider Tactics — Does a tool update need a](https://static.mm-ais.com/article-images-pixabay/does-a-tool-update-need-a-new-audit-loca-9a5d2818.jpg)

## Comparison

Compare the options side by side by lining up the live, complete commitment paths before you sign a statement of work or close an internal change ticket. When an automated employment decision tool is updated, the practical choice is usually among three options: commission a full new independent audit, commission a scoped re-audit limited to affected components, or proceed under the existing audit after documented verification. Comparing them like-for-like means totaling the same categories of work, evidence, and obligations for each path rather than comparing a vendor quote to an assumption.

| Option | Like-for-like terms to verify and total | When this option wins |
| --- | --- | --- |
| Full new independent audit | Complete audit scope, auditor independence documentation, data and model artifacts required, candidate notice and public summary deliverables, internal review and sign-off time | Wins when the update creates broad uncertainty and you need a single, defensible package before deployment |
| Scoped re-audit | Defined boundary of what is re-tested, reliance on prior audit materials, updated deliverables, and written rationale connecting scope to the change | Wins when the affected components can be cleanly isolated and prior audit evidence remains valid for everything else |
| Proceed under existing audit | Current audit coverage, change documentation, verification record, and confirmation that all required notices and summaries still align with the tool in use | Wins when documented verification shows the live tool remains within the bounds already audited and disclosed |

A full new independent audit wins when you cannot confidently bound the impact of the update across the full set of required deliverables. Before committing, verify that the proposed engagement covers every term you would otherwise need to assemble separately: the testing scope, evidence handoff responsibilities, independence attestations, and the final artifacts your organization must retain and publish. If any term is missing, the apparent option is not complete.

A scoped re-audit wins when you can produce a clear, written boundary showing exactly what changed and what did not, and the vendor or auditor can price and schedule only that work without leaving gaps. Compare the scoped proposal against the full-audit checklist item by item; the winner is the option that closes every required term with the least redundant effort, not the one with the narrowest description.

Proceeding under the existing audit wins only when your verification file is complete at the moment of commitment: the prior audit materials, the update record, and the current disclosure artifacts all match the tool you intend to use. If any element cannot be confirmed, escalate to a scoped or full audit so the committed option is complete, comparable, and defensible.

## What to do next

| Step | Action | Why it matters |
| --- | --- | --- |
| 1 | Inventory each automated employment decision tool in use and confirm an independent bias audit has been completed within the prior year. | NYC Local Law 144 requires an annual independent bias audit for every covered tool, regardless of whether patches have been applied. |
| 2 | Document the scope and nature of any recent tool update or patch, noting changes to data inputs, scoring logic, or decision outputs. | Individual patches do not automatically trigger a new audit, but understanding what changed is essential to assessing ongoing audit validity. |
| 3 | Compare the updated tool’s current functionality and use case against the scope of the existing annual bias audit on a like-for-like basis. | This comparison verifies whether the existing audit still accurately represents the tool as it is now being used. |
| 4 | If the update affects audit validity, commission a new bias audit from an independent auditor before returning the tool to use. | NYC Local Law 144 requires audits to be performed by an independent auditor, not through self-assessment, whenever the existing audit no longer represents the tool. |
| 5 | Verify the live, complete audit documentation and required disclosures are in place before committing the updated tool to active employment decisions. | Confirming the full, current compliance record before use reduces the risk of deploying a tool that falls outside the annual audit requirement. |
| 6 | Maintain a centralized record of all tool updates, validity assessments, and audit dates to track compliance with the annual audit cycle. | Clear documentation supports consistent decision-making on future updates and demonstrates ongoing adherence to NYC Local Law 144. |

## Frequently Asked Questions

**Does applying a routine patch reset the annual bias-audit clock under Local Law 144?**

Individual patches do not automatically require a new independent audit, because the annual audit cycle set by Local Law 144 does not mandate commissioning a separate audit after every tool update.

**Can an employer satisfy the rule by running its own internal review instead of hiring an outside auditor?**

Bias audits must be performed by an independent auditor, and employers using an automated employment decision tool must commission an independent audit rather than conducting a self-assessment.

**What timing window does the law use for the required audit?**

Employers must ensure an independent bias audit of each automated employment decision tool has been conducted within the prior year.

**When would a mid-year software change force an audit sooner than the annual cycle?**

Tool updates that affect audit validity may require a new audit before use, and that answer changes if a tool update alters the tool in a way that undermines the validity of the existing annual audit.

**What should an employer actually check after a software change to decide whether the old audit still covers the system?**

Employers should verify whether an update changes the tool in a way that makes the existing annual bias audit no longer representative, including whether the change altered the model, inputs, scoring logic, or decision outp.

**Where should an employer look to confirm when the bias-audit duty applies and which compliance date governs?**

Check the current official text for when the bias-audit duty applies and what compliance date governs.

## Quick answers

| Does Local Law 144 require a new bias audit after every individual patch? | Individual patches do not automatically require a new independent audit. |
| --- | --- |
| How often does Local Law 144 require an independent bias audit of an automated employment decision tool? | NYC Local Law 144 requires employers to commission an independent bias audit of an automated employment decision tool at least once per year, not after each patch. |
| Who must perform the bias audit under Local Law 144? | Bias audits must be performed by an independent auditor. |
| Can an employer conduct a self-assessment instead of commissioning an independent audit? | Employers using an automated employment decision tool must commission an independent audit rather than conducting a self-assessment. |
| When might a tool update require a new audit before use? | Tool updates that affect audit validity may require a new audit before use. |

Also worth reading: **NYC Local Law 144: $500/Day Fines, Point-in-Time Audits**: [NYC Local Law 144: $500/Day](https://ailaborbrain.com/blog/nyc-local-law-144-500day-fines-point-in-time-audits.php) · **NYC Local Law 144 Bias Audit Costs: What $1.5K Buys in 2026**: [NYC Local Law 144 Bias](https://ailaborbrain.com/blog/nyc-local-law-144-bias-audit-costs-what-15k-buys-in-2026.php) · **Which AI Hiring Tools Need a Bias Audit: NYC Local Law 144 Rules for 2026**: [Which AI Hiring Tools Need](https://ailaborbrain.com/blog/which-ai-hiring-tools-need-a-bias-audit-nyc-local-law-144-rules-for-2026.php)

### Related reading

- [NYC Local Law 144 Bias Audit Costs: How to Verify $1.5K vs $5K](https://ailaborbrain.com/blog/nyc-local-law-144-bias-audit-costs-how-to-verify-15k-vs-5k.php)
- [Resume Screening Bias: 0.80 Four-Fifths Rule vs Human Override](https://ailaborbrain.com/blog/resume-screening-bias-080-four-fifths-rule-vs-human-override.php)
- [AI job exposure thresholds: 10% vs 50% GPT-4 task-share flags](https://ailaborbrain.com/blog/ai-job-exposure-thresholds-10-vs-50-gpt-4-task-share-flags.php)
- [California AI resume screening 2026: 80% impact test to defend audits](https://ailaborbrain.com/blog/california-ai-resume-screening-2026-80-impact-test-to-defend-audits.php)
- [AI Job Risk 2026: Michael Meeks’ Task-Exposure Index — 35% vs Reskill Now](https://ailaborbrain.com/blog/ai-job-risk-2026-michael-meeks-task-exposure-index-35-vs-reskill-now.php)
- [California LWDA job funds: verify live totals before you commit](https://ailaborbrain.com/blog/california-lwda-job-funds-verify-live-totals-before-you-commit.php)

### Latest

- [Resume Screening Bias: 0.80 Four-Fifths Rule vs Human Override](https://ailaborbrain.com/blog/resume-screening-bias-080-four-fifths-rule-vs-human-override.php)
- [AI job exposure thresholds: 10% vs 50% GPT-4 task-share flags](https://ailaborbrain.com/blog/ai-job-exposure-thresholds-10-vs-50-gpt-4-task-share-flags.php)
- [California AI resume screening 2026: 80% impact test to defend audits](https://ailaborbrain.com/blog/california-ai-resume-screening-2026-80-impact-test-to-defend-audits.php)

Canonical: https://ailaborbrain.com/blog/does-a-tool-update-need-a-new-audit-local-law-144-says-1-year-not-each-patch.php
Markdown: https://ailaborbrain.com/blog/does-a-tool-update-need-a-new-audit-local-law-144-says-1-year-not-each-patch.php/index.md
