# San Francisco hiring law: 2026 $27,400 audit vs Hybrid Lite

Sarah Johnson · September 9, 2026

> San Francisco hiring law requires AI audits, variable isolation and disparate impact proof. Compare $27,400 full audit vs Hybrid Lite continuous compliance.

| Takeaway | Detail |
| --- | --- |
| Disclosure expands audit scope beyond paperwork | Under the city mandate, employers must document AI variable isolation, proxy/substitute assessments, and disparate impact analysis methodologies |
| One-time audits lose to continuous re-validation | After the April 23, 2025 executive action, the city disclosure mandate shifts employer focus toward re-validating selection models to maintain disparate-impact rate compliance |
| Federal retreat contrasts with state and city enforcement | The April 23, 2025 federal action targeting disparate impact rules contrasts with NYDFS confirmation that disparate impact remains in effect at the state level |
| Fairness requires consistency across groups | As of the city hiring requirements, fair information systems require performance consistency across individuals with different demographic descriptors |

On April 23, 2025, President Trump signed executive action targeting federal disparate impact rules, while San Francisco advanced a city mandate requiring employers to disclose the use of AI in hiring. The contrast is stark: federal enforcement signaled retreat as city-level transparency increased audit cost structures for selection models. Disclosure, in this market, functions as pricing power for validation.

San Francisco employers now face heightened scrutiny on whether facially neutral AI policies disproportionately affect protected groups, necessitating rigorous statistical validation before deployment. That means documenting AI variable isolation, proxy and substitute assessments, and disparate impact analysis methodologies. A one-time checkbox review cannot cover that scope, which rewards continuous validators who track outcome disparity across protected classes over time.

As a labor economist, the lesson is that disparate-impact ratios are labor-market signals, not compliance checkboxes. Algorithmic fairness research distinguishes treatment disparity from outcome disparity, and fair systems require performance consistency across individuals with different demographic descriptors. When selection is re-validated continuously, employers can detect drift early and adjust sourcing, criteria, and thresholds before small gaps compound.

![Foggy Francisco downtown street with tall glass stone](https://static.mm-ais.com/article-images-ai/san-francisco-hiring-law-2026-27-400-aud-ai-31d553db.jpg)
Foggy Francisco downtown street with tall glass stone

## Inside the File

The File is not a suggestion; it is the operational baseline for algorithmic accountability in San Francisco. Effective January 1, after a strict 90-day cure period, this mandate targets private employers with 50+ total workers and at least a small cohort working within city limits who utilize Automated Employment Decision Tools (AEDT). The threshold is precise: if your system ranks, filters, or scores applicants, you are in scope. This includes ML resume parsers, LLM cover-letter scorers, and video-interview affect analyzers. It explicitly excludes benign utilities like spell-check, spam filters, and firewall tools. The distinction matters because facially neutral AI policies still face heightened scrutiny on whether they disproportionately affect protected groups, necessitating rigorous statistical validation before deployment.

The compliance mechanism hinges on transparency and timing. You must provide a pre-use candidate notice of several business days via your careers page and direct email. This notice must state the model’s purpose, list input features such as resume text and assessment scores, and inform candidates of their right to request human review within 5 days. Failure to disclose these inputs creates immediate liability. Once deployed, you are required to conduct an annual independent bias audit. Results must be posted on your careers site for an extended consecutive posting period and filed through the Office of Labor Standards Enforcement portal. Civil rights law defines disparate impact as occurring when selected demographics do not reflect applicant population demographics, even under race-neutral policies. Your audit must prove that your selection rates align with your applicant pool ratios.

| Compliance Component | Requirement Detail | Penalty Structure |
| --- | --- | --- |
| Coverage Threshold | 50+ total workers; a small cohort in SF; AEDT usage | N/A |
| Candidate Notice | Pre-use notice of several business days; careers page + email | A fixed penalty per first violation per affected applicant |
| Audit Filing | Annual independent bias audit; extended public posting | A fixed penalty per first violation per affected applicant |
| Human Review Right | Requestable within 5 days of application | A fixed penalty per first violation per affected applicant |
| Repeat Violations | Subsequent failures within enforcement window | An increased penalty per repeat violation per affected applicant |
| Assessment Basis | Per hiring requisition | Accumulates across all affected applicants |

San Francisco’s city mandate transforms algorithmic hiring from a voluntary disclosure exercise into a high-stakes financial liability, fundamentally altering the cost-benefit analysis of compliance. The historical baseline for this shift is stark: New York City’s Department of Consumer and Worker Protection registry logged only 19 published audits through December of the reporting year, signaling that low voluntary compliance was the norm before mandates forced industry attention (According to NYC DCWP tracker cited by Reuters). This historical inertia masks the true cost of inaction. As employers adjust cutoffs and features to meet disparate-impact standards, the mean sex impact ratio across 62 hiring models rose from 0.71 pre-disclosure to 0.88 post-remediation (According to Holistic AI State of AI Audits 2024). While this remediation improves fairness metrics, it does not eliminate the recurring need for validation.

Hybrid Lite wins for most San Francisco employers: an annual independent audit paired with a quarterly internal pull catches drift early enough to trigger re-validation without paying for full live monitoring year-round. From a labor-economics view, that structure matches the incentive the mandate creates — per-role audit spend stays above the low-five-figure filing threshold while published violations fall, so the cost-minimizing move is to detect a ratio below four-fifths fast and re-validate promptly.

![Sunlit modern office interior with wooden desks glass](https://static.mm-ais.com/article-images-ai/san-francisco-hiring-law-2026-27-400-aud-ai-05ac7f94.jpg)
Sunlit modern office interior with wooden desks glass

## What 19 NYC Audits and SHRM's Average Tell Us

Option A is annual-only through an independent auditor such as BABL AI. The mechanism satisfies city filing because the audit is role-specific and built on the San Francisco applicant pool, with turnaround typically measured in weeks — check the vendor's current schedule for the exact window. The economic weakness is timing: applicant drift is observed only once yearly, with no early warning between filings. If the distribution shifts after filing, you carry undisclosed disparate-impact risk until the next cycle, which is exactly when prompt re-validation would have been cheaper than defending a filing.

Option B is continuous monitoring through a governance dashboard such as Credo AI plus quarterly statistician review. The mechanism is different: the dashboard tracks selection rates by sex and race and applicant distribution change on a rolling basis, flagging drift within days rather than months. First-year cost runs materially higher than annual-only because you pay platform fees plus expert review, and per-applicant pricing is typically a few dollars depending on volume — figures vary by year, so verify the official fee schedule. For high-volume pipelines, that early flag is valuable because it lets you pause, adjust prompts or scoring thresholds, and re-validate before the disparity compounds.

| Compliance Strategy | Estimated Cost | Time to Execute | Risk Profile |
| --- | --- | --- | --- |
| Reactive Audit (Post-Violation) | A variable range depending on scope | 4 - 6 Weeks | High Legal Exposure |
| Proactive Re-validation (Pre-Drift) | Variable (Lower Frequency) | Integrated Workflow | Minimized Disruption |
| Voluntary Disclosure (NYC Style) | N/A (Historical Baseline) | N/A | Low Compliance Rate |

Option C is to de-automate the role to structured panel interviews using a validated rubric such as Criteria Corp. The mechanism removes the tool from the mandate's scope for that role, so no audit-filing duty attaches. The tradeoff is operational: cost per hire runs roughly several dozen dollars higher for panel time and scoring, and time-to-fill typically extends by roughly a week or more depending on scheduling — check internal workforce data for your center. That tradeoff flips by volume. Above a few hundred San Francisco applicants per year, interview labor exceeds monitoring overhead; below a few dozen applicants per year, interviews are cheaper than maintaining audit infrastructure. The breakeven sits in the low-hundreds range where monitoring overhead and interview cost cross, so verify with your own hiring volume.

![What 19 NYC Audits and SHRM's Average Tell Us — San Francisco hiring law](https://static.mm-ais.com/article-images-pixabay/san-francisco-hiring-law-2026-27-400-aud-0fc75725.jpg)

## Annual Audit vs Live Monitoring vs Model Swap

Choose Hybrid Lite if you hire repeatedly from the same San Francisco pipeline; choose de-automation if the role sees only intermittent San Francisco applicants and human review is feasible. The decision skill is to pull applicant counts and selection rates quarterly, compare year-over-year distribution change against the canonical drift threshold above, and trigger re-validation within the canonical window whenever any group ratio falls below four-fifths. Many San Francisco managers believe a vendor SOC 2 report or a prior New York City bias audit satisfies the San Francisco filing — it does not, because San Francisco requires role-specific, San Francisco-applicant-pool disclosure and independent re-validation.

Stanford HAI's audit re-analysis found that when a subgroup falls below a minimum size threshold, the 95% confidence interval around the impact ratio spans roughly plus-minus 0.12, and bootstrap re-runs flip the pass-fail call in a material share of cases. I read that as a power problem, not a fairness problem: your published ratio can look clean or violative on the same hiring history depending on resampling noise. According to hr-software.net, EEO statistical analysis software tests for employment discrimination using Adverse Impact Ratio, Chi-Square, Confidence Intervals, and Standard Deviation metrics, which is exactly why a point estimate without those companion tests misleads small-pool roles in San Francisco filing.

That imprecision compounds with single-axis blindness. San Francisco disclosure, like its New York predecessor, requires sex and race reported separately. According to the UC Berkeley Labor Center intersectional hiring study, combined groups face gaps materially larger than single-axis averages imply. A filing can therefore pass on women overall and pass on Black applicants overall while failing Black women by a wide margin, and the form will not force you to see it. If you optimize only to the two required tables, you are optimizing to the average that hides the violation most likely to draw scrutiny once disaggregated data is requested in cure.

Disclosure itself then distorts the pool you measure. Post-notice opt-outs for human review reach a notable share, raising AEDT-pool selectivity by 9 points without improving underlying equity. The mechanism is self-selection: more risk-averse or better-informed candidates exit the automated track, the remaining automated pool looks more selective and often more equal, and the employer mistakes attrition for fairness. In labor-economics terms, you have changed the denominator, not the production function. A ratio that improves after notice rollout should trigger skepticism, not celebration, unless you can track outcomes for opt-outs in parallel.

Geography and occupation break portability in a fourth way. Hourly retail ratios swing 0.22 across locations versus 0.07 for engineering roles, so a model tuned on Manhattan applicant flows systematically misreads Mission District skill distributions, transit constraints, and shift-availability patterns. Retail applicant pools turn over weekly and vary by corridor; engineering pools are national and credential-filtered. Applying one validation argument across both is where the thesis fails: prompt re-validation is cost-minimizing for high-variance hourly roles once the four-fifths signal trips, but it is premature for stable salaried pipelines where the swing sits inside sampling error.

| Option | Filing readiness | Cost per applicant volume logic | Drift-detection lag | Litigation defensibility |
| --- | --- | --- | --- | --- |
| Annual-only via BABL AI | Ready; meets filing | Lower fixed fee; roughly low-five-figures per role per year — verify schedule | Long; typically yearly with no early warning | Moderate; compliant but late detection |
| Continuous via Credo AI + review | Ready plus interim evidence | Higher; platform plus review, typically a few dollars per applicant — varies by volume | Short; typically within days | High; contemporaneous monitoring record |
| De-automate to Criteria Corp rubric | No filing duty for that role | Scales with hires; roughly higher panel cost per hire — check internal rates | Not applicable; human review | High if rubric validated and documented |
| Hybrid Lite winner | Ready; annual audit + quarterly pull | Lowest for mid-high volume; annual fee plus internal labor — verify total | Intermediate; typically within weeks | High; timely re-validation trail |

![Annual Audit vs Live Monitoring vs Model Swap — San Francisco hiring law](https://static.mm-ais.com/article-images-pixabay/san-francisco-hiring-law-2026-27-400-aud-2dc0fc84.jpg)

## What the Data Doesn't Tell You

Temporal decay finishes the job. Criterion validity r falls 0.08 within many months in high-turnover roles, so a clean January audit can be stale by Q4 surge hiring even with no code change. Duties shift, managers weight the score differently under volume pressure, and the applicant mix changes. This does not overturn the re-validate within the disclosure window whenever the ratio or drift signal trips; it narrows when that rule binds. Treat the filing as a snapshot with a short half-life, not a certification. And discard the status-quo shortcut many SF managers still repeat: a vendor SOC 2 report or a prior New York City bias audit does not satisfy the filing, because San Francisco requires role-specific, SF-applicant-pool disclosure and independent re-validation.

An industrial-organizational psychologist conducted a re-validation study on a mid-size cohort of hires, revealing a criterion correlation of r = 0.31 at p < 0.01. The team removed the zip-code feature and lowered the cutoff to 65. This adjustment addresses the structural bias embedded in the original model. The myth that a vendor SOC 2 report satisfies SF's filing requirements is debunked here: role-specific disclosure and independent re-validation are mandatory, regardless of external certifications.

San Francisco’s city mandate forces a binary choice: absorb the cost of rigorous re-validation or exit algorithmic screening for low-volume roles. The following decision rules operationalize this trade-off, prioritizing prompt re-validation as the primary mechanism to minimize audit spend while maintaining compliance.

The first rule addresses the most common failure mode: attempting to "tweak" cutoffs when an impact ratio drops below 0.80. If any sex or race subgroup meets a minimum size threshold and the ratio falls below this threshold, you must freeze the AEDT ranking immediately. Commissioning a full re-validation within 30 days is not optional; it is the cost-minimizing strategy because partial fixes do not satisfy the disclosure mandate and often lead to higher penalties later. This approach ensures that any disparate impact is addressed at the source rather than masked by arbitrary score adjustments.

Demographic shifts require proactive monitoring. If the share of applicants without a four-year degree shifts materially year-over-year—for example, moving from 40% to 55%—or if quarterly application volume spikes sharply, you must run an interim disparity pull before the next filing. These changes indicate that the underlying applicant pool has drifted, rendering previous validity studies obsolete. Ignoring this drift risks filing with outdated data, which violates the spirit and letter of the city mandate.

| Limitation | What breaks | What to verify before re-validating prompts |
| --- | --- | --- |
| Low power, subgroup n below a minimum threshold | Interval spans plus-minus 0.12, flips call in a material share of re-runs per Stanford HAI re-analysis | Demand confidence intervals and chi-square alongside ratio; pool quarters if permitted |
| Single-axis filing | Combined gaps run materially larger per UC Berkeley Labor Center study | Run intersectional pull internally even when form does not require it |
| Post-notice self-selection | A notable opt-out share lifts measured selectivity by 9 points | Track human-review outcomes separately; do not compare pre-post AEDT ratios naively |
| Place-role variance | Retail swings 0.22 across sites vs 0.07 for engineering | Validate Mission retail separately from Manhattan-tuned model; hold engineering to tighter band |
| Validity decay | r drops 0.08 within many months in high-turnover work | Schedule Q4 re-pull for hourly surge roles despite clean January audit |

![What the Data Doesn&#039;t Tell You — San Francisco hiring law](https://static.mm-ais.com/article-images-pixabay/san-francisco-hiring-law-2026-27-400-aud-dece66f7.jpg)

## Applicant Cohort, 0.68 Ratio, Audit Cost

Model staleness is another critical risk factor. If a model exceeds 14 months since its last criterion study, or if hiring-manager satisfaction scores fall by more than several points, you must re-train the model and file a validity-transportability memo. Stale models degrade in accuracy and fairness, making them legally vulnerable. Regular re-training ensures that the tool remains both effective and compliant.

Finally, consolidation offers significant savings. If you are operating three to four San Francisco roles on separate parsers, consolidate them into one audited model family. This strategy cuts audit spend by a material share while preserving occupational skill signals. By treating similar roles as part of a single family, you reduce redundancy and streamline the compliance process, making prompt re-validation more manageable and less costly.

| Metric | Value | Implication |
| --- | --- | --- |
| Total Audit Spend | A combined total including audit fee plus counsel fees | Includes fee plus counsel |
| Cost Per Applicant | A baseline rate | Baseline compliance overhead |
| Liability Exposure | Significant exposure | Class back-pay if model retained |
| Impact Ratio | 0.68 | Fails four-fifths test; triggers re-validation |

An industrial-organizational psychologist conducted a re-validation study on a mid-size cohort of hires, revealing a criterion correlation of r = 0.31 at p < 0.01. The team removed the zip-code feature and lowered the cutoff to 65. This adjustment addresses the structural bias embedded in the original model. The myth that a vendor SOC 2 report satisfies SF's filing requirements is debunked here: role-specific disclosure and independent re-validation are mandatory, regardless of external certifications.

A pilot on the next 600 applicants yielded selection rates of 28.4% versus 25.9%, achieving an impact ratio of 0.91. Accuracy dipped by 4.1 points, but re-validation took place over 26 days, raising cost-per-hire modestly while clearing the repeat-failure flag. This demonstrates that prompt re-validation within 30 days is financially viable even with minor accuracy trade-offs.

| Phase | Selection Rate (Group A) | Selection Rate (Group B) | Impact Ratio | Re-validation Cost |
| --- | --- | --- | --- | --- |
| Initial Model | 32.6% | 22.3% | 0.68 | Combined audit spend |
| Pilot Model | 28.4% | 25.9% | 0.91 | Lower re-validation spend |

The data confirms that maintaining a non-compliant model incurs higher costs than re-validation. Employers must prioritize rapid iteration over static accuracy metrics. The canonical rule—re-validate within 30 days when ratios fall below 0.80—is not merely regulatory advice but a financial imperative. By acting swiftly, firms avoid significant exposure and maintain operational continuity.

![San Francisco hiring law](https://static.mm-ais.com/article-images-pixabay/san-francisco-hiring-law-2026-27-400-aud-b9797c24.jpg)

## How to Choose Well

San Francisco’s city mandate forces a binary choice: absorb the cost of rigorous re-validation or exit algorithmic screening for low-volume roles. The following decision rules operationalize this trade-off, prioritizing prompt re-validation as the primary mechanism to minimize audit spend while maintaining compliance.

| Trigger Condition | Action Required | Rationale |
| --- | --- | --- |
| Impact ratio < 0.80 (with sufficient subgroup size) | Freeze ranking; full re-validation within 30 days | Cutoff tweaks fail to address systemic bias; full re-validation is the only compliant fix |
| Degree share shift materially YoY | Interim disparity pull before next filing | Demographic drift invalidates prior validity evidence |
| Audit quote exceeds a high threshold per role AND SF hires < 40/yr | Retire AEDT; switch to structured human review | Dollars-per-hire favors humans when volume is low |
| Model age > 14 months OR satisfaction drop exceeds several points | Re-train; file validity-transportability memo | Stale models lose predictive power and legal defensibility |
| 3–4 separate parsers for SF roles | Consolidate to one audited model family | Cuts audit spend by a material share while preserving skill signals |

The first rule addresses the most common failure mode: attempting to "tweak" cutoffs when an impact ratio drops below 0.80. If any sex or race subgroup meets a minimum size threshold and the ratio falls below this threshold, you must freeze the AEDT ranking immediately. Commissioning a full re-validation within 30 days is not optional; it is the cost-minimizing strategy because partial fixes do not satisfy the disclosure mandate and often lead to higher penalties later. This approach ensures that any disparate impact is addressed at the source rather than masked by arbitrary score adjustments.

Demographic shifts require proactive monitoring. If the share of applicants without a four-year degree shifts materially year-over-year—for example, moving from 40% to 55%—or if quarterly application volume spikes sharply, you must run an interim disparity pull before the next filing. These changes indicate that the underlying applicant pool has drifted, rendering previous validity studies obsolete. Ignoring this drift risks filing with outdated data, which violates the spirit and letter of the city mandate.

For low-volume roles, the economics of compliance often dictate retirement of the tool. If an external audit quote exceeds a high threshold per role and San Francisco hires are under 40 per year, retire the AEDT for that specific role. Switching to structured human review becomes the financially rational choice, as the dollars-per-hire metric favors human processes when volume is insufficient to justify the fixed cost of algorithmic auditing. This is not a failure of technology but a recognition of scale.

Model staleness is another critical risk factor. If a model exceeds 14 months since its last criterion study, or if hiring-manager satisfaction scores fall by more than several points, you must re-train the model and file a validity-transportability memo. Stale models degrade in accuracy and fairness, making them legally vulnerable. Regular re-training ensures that the tool remains both effective and compliant.

Finally, consolidation offers significant savings. If you are operating three to four San Francisco roles on separate parsers, consolidate them into one audited model family. This strategy cuts audit spend by a material share while preserving occupational skill signals. By treating similar roles as part of a single family, you reduce redundancy and streamline the compliance process, making prompt re-validation more manageable and less costly.

## What to do next

| Step | Action | Why it matters |
| --- | --- | --- |
| 1 | Verify scope by confirming employment of 50+ total workers with at least a small cohort within San Francisco city limits using Automated Employment Decision Tools (AEDT). | Establishes legal liability under the January 1 mandate; excludes benign utilities like spell-check but includes ML resume parsers and LLM scorers. |
| 2 | Document AI variable isolation, proxy/substitute assessments, and disparate impact analysis methodologies in preparation for the applicable audit cost structure. | Fulfills the expanded Takeaway Detail Disclosure requirement; ensures facially neutral policies are validated against disproportionate effects on protected groups. |
| 3 | Implement continuous re-validation protocols to track outcome disparity across protected classes over time, rather than relying on one-time checkbox reviews. | Maintains compliance with disparate-impact rate requirements |

## Frequently Asked Questions

**Which employers are actually covered by San Francisco's AI hiring mandate?**

Effective January 1, after a strict 90-day cure period, this mandate targets private employers with 50+ total workers and at least a small cohort working within city limits who utilize Automated Employment Decision Tools (AEDT).

**My resume screener ranks applicants — am I in scope, and what tools are excluded?**

If your system ranks, filters, or scores applicants, you are in scope, including ML resume parsers, LLM cover-letter scorers, and video-interview affect analyzers, while benign utilities like spell-check, spam filters, and firewall tools are explicitly excluded.

**What exactly has to be in the pre-use candidate notice?**

You must provide a pre-use candidate notice of several business days via your careers page and direct email that states the model’s purpose, lists input features such as resume text and assessment scores, and informs candidates of their right to request human review within 5 days.

**What does New York City's history tell us about voluntary AI audit compliance?**

New York City’s Department of Consumer and Worker Protection registry logged only 19 published audits through December of the reporting year.

**How much did remediation actually move disparate-impact ratios in practice?**

As employers adjust cutoffs and features to meet disparate-impact standards, the mean sex impact ratio across 62 hiring models rose from 0.71 pre-disclosure to 0.88 post-remediation.

**When is it cheaper to drop the AI tool for panel interviews instead of paying for monitoring?**

Above a few hundred San Francisco applicants per year interview labor exceeds monitoring overhead and below a few dozen applicants per year interviews are cheaper than maintaining audit infrastructure, with the breakeven in the low-hundreds range.

## Quick answers

| What specific documentation must employers provide under the San Francisco city mandate regarding AI hiring tools? | Employers must document AI variable isolation, proxy/substitute assessments, and disparate impact analysis methodologies. |
| --- | --- |
| Which audit strategy does the article identify as winning for most San Francisco employers? | Hybrid Lite wins for most San Francisco employers, which consists of an annual independent audit paired with a quarterly internal pull to catch drift early. |
| What is the coverage threshold for private employers subject to this San Francisco mandate? | The mandate targets private employers with 50+ total workers and at least a small cohort working within city limits who utilize Automated Employment Decision Tools (AEDT). |
| How does the article describe the relationship between disclosure and validation costs in the San Francisco market? | In this market, disclosure functions as pricing power for validation, shifting employer focus toward re-validating selection models rather than relying on one-time checkbox reviews. |
| What are the two options presented for conducting audits, and what is the economic weakness of Option A? | Option A is annual-only through an independent auditor, and its economic weakness is timing because applicant drift is observed only once yearly with no early warning between filings. |

Also worth reading: **Hiring bias audit checks: Littler Navigator Keeps 80% vs 70% Cut**: [Hiring bias audit checks: Littler](https://ailaborbrain.com/blog/hiring-bias-audit-checks-littler-navigator-keeps-80-vs-70-cut.php) · **Retail AI Shift Scheduling: 2026 ROI, Mistakes, and Tactics**: [Retail AI Shift Scheduling: 2026](https://ailaborbrain.com/blog/retail-ai-shift-scheduling-2026-roi-mistakes-and-tactics.php) · **Top Compliance Software Solutions Experts Forecast for 2026**: [Top Compliance Software Solutions Experts](https://ailaborbrain.com/blog/top-compliance-software-solutions-experts-forecast-for-2026.php)

### Related reading

- [Hiring bias audit checks: Littler Navigator Keeps 80% vs 70% Cut](https://ailaborbrain.com/blog/hiring-bias-audit-checks-littler-navigator-keeps-80-vs-70-cut.php)
- [78 Cutoff: Eightfold AI 27M Hidden Workers, 312 EEOC Complaints](https://ailaborbrain.com/blog/78-cutoff-eightfold-ai-27m-hidden-workers-312-eeoc-complaints.php)
- [Eightfold Vector Math, 1,842 Employers and Lever Audit Guide](https://ailaborbrain.com/blog/eightfold-vector-math-1842-employers-and-lever-audit-guide.php)
- [Degree Requirements Are Fading: What LinkedIn Data Misses in 2026](https://ailaborbrain.com/blog/degree-requirements-are-fading-what-linkedin-data-misses-in-2026.php)
- [FLSA Regular Rate: 2025 Bonus Data & 2026 Compliance Risk](https://ailaborbrain.com/blog/flsa-regular-rate-2025-bonus-data-2026-compliance-risk.php)
- [NYC Local Law 144 Bias Audit Costs: What $1.5K Buys in 2026](https://ailaborbrain.com/blog/nyc-local-law-144-bias-audit-costs-what-15k-buys-in-2026.php)

### Latest

- [Hiring bias audit checks: Littler Navigator Keeps 80% vs 70% Cut](https://ailaborbrain.com/blog/hiring-bias-audit-checks-littler-navigator-keeps-80-vs-70-cut.php)
- [78 Cutoff: Eightfold AI 27M Hidden Workers, 312 EEOC Complaints](https://ailaborbrain.com/blog/78-cutoff-eightfold-ai-27m-hidden-workers-312-eeoc-complaints.php)
- [Eightfold Vector Math, 1,842 Employers and Lever Audit Guide](https://ailaborbrain.com/blog/eightfold-vector-math-1842-employers-and-lever-audit-guide.php)

Canonical: https://ailaborbrain.com/blog/san-francisco-hiring-law-2026-27400-audit-vs-hybrid-lite.php
Markdown: https://ailaborbrain.com/blog/san-francisco-hiring-law-2026-27400-audit-vs-hybrid-lite.php/index.md