Arena Analytics

In the abstract of a 2020 MLconf talk, Arena's then Chief Data Scientist (now CTO) describes 'Adversarial Fairness': a KPI-predicting network is trained against a discriminator that tries to recover EEOC protected classes until they reach Nash equilibrium, and the method is said to be demonstrated in a scaled production environment.

https://mlconf.com/sessions/adversarial-fairness-in-the-labor-market/

The Retention Prediction page says Arena uses 'scores of data points specific to each role and organization' and combines proprietary, external and applicant data with a proprietary bias-mitigation model, but names no features, model type, validation results or limitations.

https://arena.io/retention-prediction

A January 2021 release announces an external AI Ethics Advisory Board chaired by HR-tech analyst John Sumser, but no outputs from the board could be found and its page on arena.io now returns 404.

https://www.prnewswire.com/news-releases/leading-hr-technology-company-launches-ai-ethics-advisory-board-301215510.html

No model card, system card, technical documentation pack, explainability statement, AI policy or ISO 42001 reference exists on arena.io or in the help.arena.io help center; all 19 sitemap URLs are marketing or legal pages, and /security, /trust, /responsible-ai and /ethics return 404.

Have a URL that disputes this?

The homepage says Arena removes 'protected characteristics and related signals' from its models and that its recommended candidate group 'consistently exceeds the EEOC's 80% compliance threshold, often surpassing 95%', with no auditor, date, sample or impact-ratio tables.

https://arena.io

A November 2020 release claims Arena's adversarial technique removes '92%-99% of latent bias' and lifts representation above 95%, based on Arena's own demographic-parity measurements rather than an independent audit.

https://www.prnewswire.com/news-releases/machine-learning-removes-bias-from-algorithms-and-the-hiring-process-301167669.html

We found no NYC LL 144 bias-audit summary or independent audit on arena.io, help.arena.io, the ACLU's crowd-sourced LL 144 audit tracker or Warden AI (trust.warden-ai.com/arena returns 404), and no BABL AI, DCI, ORCAA, Holistic AI or Credo AI audit of Arena.

Have a URL that disputes this?

We found no EU AI Act material, Article 27 FRIA template, deployer-obligation guidance, intended-use statement or system card on arena.io, in its 27-article help center or in its press releases.

Have a URL that disputes this?

The privacy policy (updated March 30, 2022) says the services are controlled and operated from the United States and warns users outside the US that their data may be stored in a country EU regulators do not deem an 'adequate jurisdiction', without mentioning the EU AI Act or deployer duties.

https://arena.io/privacy

The privacy policy lists the applicant data Arena collects (resume, background, question responses, on-platform interactions, device usage and information 'in the public domain'), says demographic data is used only to validate bias reduction and client-supplied hire and termination dates only to train the algorithms, and sets no retention period beyond 'as long as reasonably necessary'.

https://arena.io/privacy

The homepage says Arena's dataset covers 4.2 million people and their career outcomes, growing by 70,000 a month, and that protected characteristics and related signals are removed from its models.

https://arena.io

Arena's founder said adversarial debiasing let the company start using data 'once considered unusable, such as commuting distance', a location-derived input that can act as a proxy for protected classes.

https://www.prnewswire.com/news-releases/machine-learning-removes-bias-from-algorithms-and-the-hiring-process-301167669.html

The candidate FAQ says Arena collects applicant 'responses, text, and interactions' on its platform to generate a prediction for each candidate in each position, with no feature list or exclusion list.

https://help.arena.io/api/v2/help_center/en-us/articles/115004279747.json

The Why Arena page claims the platform is 'SOC 2, type II compliant', but no report, bridge letter or trust portal is published.

https://arena.io/why-arena

The only public record of an actual examination is an April 2021 release announcing a SOC 2 Type I examination by the CPA firm A-LIGN under the AICPA security principle.

https://www.prnewswire.com/news-releases/arena-analytics-completes-soc-2-type-1-examination-301267466.html

Arena publishes no training-data provenance statement, detailed exclusion list, subgroup performance metrics, subprocessor list, DPA or retention schedule.

Have a URL that disputes this?

The Talent Discovery page says Arena scores all interested candidates and 'will only recommend people who scored as likely to be retained', so others are filtered out before any employer sees them.

https://arena.io/talent-discovery

The homepage says Retention Prediction flags first-day risk, prioritises candidates 'statistically more likely to stay' before offers are made and reroutes others to different openings, and describes no override controls, score explanations or audit logs.

https://arena.io

A 2018 customer account says hiring managers received the retention prediction as 'an additional piece of information' alongside the application, and people conducted the interviews and made the final hiring decision.

https://www.prnewswire.com/news-releases/benchmark-major-northeast-senior-living-provider-reports-early-success-with-arenas-artificial-intelligence-solution-for-hiring-staff-300741471.html

In 2019 Arena's then-president said algorithms 'don't replace' human judgment and 'you can't have an algorithm make a hire'.

https://www.prnewswire.com/news-releases/artificial-intelligence-helps-to-overcome-many-of-the-biases-in-hiring-but-doesnt-replace-the-human-factor-arenas-norton-and-others-tell-a-big-crowd-at-south-by-southwest-conference-300812537.html

The candidate FAQ says applicants cannot see their results 'in accordance to company policy' and that the score is about retention, not intelligence, skill or experience.

https://help.arena.io/api/v2/help_center/en-us/articles/203883878.json

The candidate FAQ says completing the questionnaire is not obligatory but that an application 'may not be considered complete' without it, and it describes no alternative selection process.

https://help.arena.io/api/v2/help_center/en-us/articles/115004278567.json

We found no public documentation of recruiter override controls, confidence thresholds, in-product score explanations, per-jurisdiction settings or decision audit logs.

Have a URL that disputes this?

The Retention Prediction page says post-hire reports 'clearly lay out the retention prediction numbers' to back its guarantee of greater retention for 'Likely' hires; this is customer-facing outcome tracking, not public or fairness monitoring.

https://arena.io/retention-prediction

The candidate help center gives a support email and toll-free number for questionnaire access problems, the only public support channel besides legal@arena.io.

https://help.arena.io/api/v2/help_center/en-us/articles/115004427928.json

We found no working status page (status.arena.io points to Atlassian Statuspage but serves no page), no security.txt or vulnerability-disclosure contact, no incident history and no model-update changelog.

Have a URL that disputes this?

The public candidate help center has 27 FAQ articles covering the 10-15 minute questionnaire, supported devices, six-month result validity and support contacts, but nothing for employers on bias, notices or accommodations.

https://help.arena.io/api/v2/help_center/en-us/articles.json?per_page=100

The Retention Prediction page lists ATS/HRIS integrations with Workday, ADP, UKG, Paycom, iCIMS, Talent Reef, Jobvite, Oracle and Symplr.

https://arena.io/retention-prediction

Product pages such as Internal Mobility describe each product in a few marketing lines (for example, employees searching 'a curated list of open jobs within your organization where they are likely to be retained') with no setup, configuration or compliance guidance.

https://arena.io/internal-mobility

Arena publishes no resource hub, whitepapers, deployer or admin guide, DPA, or guidance on NYC LL 144, Illinois, Colorado, California or the EU AI Act; /blog and /news redirect to a /resources page that returns 404.

Have a URL that disputes this?