AI Use Policy
How Atlas uses AI, what it does not do, and how customers should use its outputs. Last reviewed: June 2026.
Atlas Standing Posture
Atlas is decision-support intelligence. Its outputs surface patterns from evidence the customer has authorized Atlas to read. Atlas does not, and is not designed to, replace human judgment in any employment, retention, compensation, promotion, or other consequential decision about a named individual.
Atlas has not been third-party audited for protected-class bias as of this writing. A bias-audit infrastructure is in active development (Q3 2026 target). Customers operating in jurisdictions requiring formal bias-audit reports (e.g., New York City) should treat Atlas outputs as one input to a human-led decision process, not as the basis for an automated employment decision.
1. What Atlas does
Pattern-matching on customer-authorized evidence
Atlas reads data the customer has explicitly connected (HRIS, CRM, ATS, finance, document uploads) plus public-web evidence the customer has scoped (LinkedIn profiles, licensing-board registries, public project documents). Atlas surfaces patterns — capability overlaps, gaps, concentration risks, financial impact — that would take a human analyst days to find manually.
Source-cited reasoning
Every Atlas claim about an employee, opportunity, or workforce decision is tied to a specific evidence source the customer can re-read. If Atlas cannot cite a specific skill, certification, project, or document, it does not make the claim.
Aggregation and ranking
Atlas counts, intersects, groups, and ranks the data customers authorize. Examples: "employees with both AICP and CEQA experience in the Sacramento office," "disciplines where coverage is concentrated in fewer than three people," "opportunities at risk if staffing gaps persist."
Documented business-risk framing
Atlas surfaces business risks (knowledge concentration, geographic concentration, single points of failure) tied to workforce structure. It frames these strictly as organizational risks the firm faces, not as life-stage predictions about individuals.
2. What Atlas never does
The following patterns are codified prohibitions inside Atlas's engine and prompts. They have been reviewed by an external Industrial-Organizational psychologist and an employment-law specialist as material exposure if implemented.
✕Turnover or flight-risk prediction
Atlas does not predict who will leave. The underlying signals (tenure, comp benchmarks, role progression) exist in customer data but are never composed into a predictive score about a named individual. Such prediction is unfalsifiable and creates legal discovery risk when used for retention or layoff decisions.
✕Performance prediction
Atlas does not predict who will become a top performer. Such inference is unfalsifiable and biased by the training assumptions of any model that produces it.
✕Burnout, wellness, or mental-state inference
Atlas does not infer or report on individual wellness, stress, or burnout. These signals are invasive, unsupported by science, and brand-damaging if surfaced.
✕Culture-fit or personality matching
Atlas does not produce 'vibe match' or culture-fit scores between candidates and roles. Such inferences rely on protected-class proxies and disparate-impact risk.
✕Age, retirement-timing, or life-stage inference
Atlas does not frame any output as 'approaching retirement,' 'near end of career,' or any equivalent. The legitimate underlying signal (long tenure plus business risk of knowledge loss) is surfaced strictly as 'deep institutional knowledge' or 'hard-to-replace tenure' — never as an age inference. ADEA exposure is a hard line.
✕Protected-class inference
Atlas does not infer gender, race, disability, national origin, religion, or any other protected attribute from names, photos, schools, or other proxies. Atlas outputs are filtered for protected-class terms on both input (queries) and output (results).
3. How customers should use Atlas outputs
Decision-support, not decision-determination
Atlas outputs are intended to inform human decision-makers — recruiters, hiring managers, COOs, BD leaders, finance officers. They are not designed to be the sole basis for employment decisions about a named individual. Customers using Atlas in employment contexts (hiring, promotion, retention, compensation, layoff) must apply human judgment before acting on any Atlas recommendation.
Evidence chains are auditable
Every Atlas claim about an individual includes the source evidence (skill name, certification, project, office, document URL). When defending an employment decision informed by Atlas, the customer can trace the underlying evidence and apply their own judgment. If an Atlas claim is not auditable to specific source evidence, it is a bug; please report to trust@atlas.ai.
Recommendations include alternatives
When Atlas recommends a staffing action (hire, promote, backfill, redeploy), it explicitly surfaces the alternatives it evaluated and why the recommended action was scored higher. Customers can interrogate these alternative paths and choose differently. (Note: the alternatives ladder is in active rollout as of June 2026.)
4. Bias-audit and AEDT compliance posture
Bias audit (current state)
Atlas has not been third-party audited for protected-class bias. The matching engine's weights are author-asserted (derived from Ken Duggan's 15+ years of AEC recruiting experience) and have not been validated against a calibration cohort. A bias-audit infrastructure capturing scoring-output disparate-impact across protected-class proxies is in active development.
NYC Local Law 144 (AEDT)
Customers in New York City using Atlas in an automated employment decision capacity must perform their own independent bias audit and provide candidates with the notice required under NYC LL 144. Atlas's current posture does not satisfy AEDT requirements on its own. We are tracking against Q3 2026 for Atlas-generated bias-audit reports.
Colorado AI Act / Illinois HB 3773
This page constitutes Atlas's consumer-facing AI use policy as required by Colorado's AI Act and Illinois HB 3773. Atlas operates as a system provider; customers using Atlas outputs in consequential employment decisions are deployers and bear deployer obligations under each statute.
EU AI Act (Article 6 / Article 27)
Workforce-management AI is classified as high-risk under EU AI Act Annex III. Atlas is currently deployed in the United States. Customers in the EU/UK should contact trust@atlas.ai to discuss fundamental-rights impact assessment, transparency obligations, and risk-management system requirements before deploying Atlas in EU/UK consequential-decision contexts.
5. Data subject rights
Access, correction, deletion
Individuals whose data appears in Atlas may request access, correction, or deletion of that data. For Atlas-discovered evidence (public-web-sourced), requests are processed at privacy@atlas.ai. For customer-supplied evidence (employee data uploaded by an Atlas customer), requests must go to the customer as data controller; Atlas processes such requests within 30 days when the customer authorizes them.
Public-web evidence (employees / candidates)
Atlas reads publicly available web content (LinkedIn profiles, licensing-board registries, conference rosters, firm team pages) and stores excerpts as enrichment evidence with source URLs. Individuals can request removal of specific source citations or full evidence records at privacy@atlas.ai.
EU / UK / California data subjects
Atlas honors GDPR Article 15 (access) and Article 17 (erasure) requests, and California CCPA/CPRA equivalents, within statutory deadlines. Endpoint-based self-service DSAR is in active development.
6. Updates to this policy
Versioning
This policy is reviewed every quarter and on every material change to Atlas's AI surfaces. Material changes include: new prediction features (which we do not plan to introduce), new bias-audit results, new compliance certifications, or scope changes to what Atlas reads.
Contact
Questions, concerns, or compliance reviews: trust@atlas.ai. Data-subject rights: privacy@atlas.ai.
Atlas is operated by Talent Phantom, Inc., a Delaware C-corp. Headquartered in the United States. This policy is provided in good faith and represents Atlas's current operating posture; it does not constitute legal advice and is not a substitute for a customer's own compliance review.