Adaface
HireAIScore rates Adaface 39 out of 100 (grade F, Substantial gaps) under rubric v1.0, last reviewed September 25, 2026. Adaface ranks 67 of 82 vendors rated in Assessment AI, against a category average of 46. That score is drawn from 31 evidence items across 7 rubric criteria for Adaface — 25 with a cited source, 6 recording that nothing was located.
Score breakdown
Adaface, criterion by criterion
Each rubric criterion is scored out of 100 on the cited evidence, then weighted into the total.
Weighted contribution shown to the right of each bar.
Article 11 Technical Documentation
Weight 20% · category avg 50
38
+7.6
Bias Audit Transparency
Weight 18% · category avg 41
22
+4.0
FRIA Support
Weight 15% · category avg 31
30
+4.5
Data Governance Disclosure
Weight 15% · category avg 53
52
+7.8
Human Oversight Design
Weight 12% · category avg 55
50
+6.0
Post-Market Monitoring
Weight 12% · category avg 38
38
+4.6
Customer Documentation
Weight 8% · category avg 58
58
+4.6
Total weighted score
39.1(rounded to 39 for display)
Category avg is the mean raw score on that criterion across the 82 Assessment AI vendors in scope of this rubric, this one included.
Strongest criterion
Customer Documentation
58raw / 100
Contributes 4.6 to the total. 58 against a 58 category average.
Weakest criterion
Bias Audit Transparency
22raw / 100
Contributes 4.0 to the total. 22 against a 41 category average.
Cited evidence
The evidence behind the score
Every score is backed by at least one cited piece of evidence. Open a criterion to see what was found, when, and where.
Download diligence recordEvidence ledger
- Items
- 31
- Documentation
- 22
- Public statement
- 3
- Absence
- 6
- With a source URL
- 25 of 31
- Source hostnames
- 3
- Fewest items
- 2
- FRIA Support
These figures measure how thoroughly Adaface was reviewed, not how Adaface performed — an absence row, recording that nothing was located, is counted like any other item.
Article 11 Technical Documentation
4 items38
- DocumentationCaptured Sep 25, 2026
Describes a question-design process led by subject-matter experts, with topics ranked for a 'minimally qualified candidate', peer review, and error checks for clarity and cultural sensitivity; it also requires questions to avoid 'gender, cultural, and vernacular bias'.
- DocumentationCaptured Sep 25, 2026
Explains deterministic coding scoring: prorated credit for each hidden test case passed, with the best submission counting, and deductions of up to 40% for hints and 25% for extra time.
https://help.adaface.com/en/articles/6886300-how-are-coding-questions-scored
- DocumentationCaptured Sep 25, 2026
The DPA's sub-processor list names OpenAI, Anthropic and Google (Gemini API) for 'large language model processing for assessment authoring and scoring'; this is the only public disclosure that LLMs score candidates.
- AbsenceCaptured Sep 25, 2026
We searched the Trust Guide, the Science hub, the 2,587-URL sitemap and the 49-article help center and found no system or model card, explainability statement, AI policy, instructions for use or ISO 42001; the Science page cites general Schmidt and Hunter validity figures, not Adaface validation data.
Bias Audit Transparency
3 items22
- AbsenceCaptured Sep 25, 2026
No NYC LL 144 bias-audit summary, independent fairness audit (BABL AI, DCI, ORCAA, Warden AI, Holistic AI, Credo AI) or adverse-impact data was found on adaface.com, in its sitemap or help center, or in web searches, and trust.warden-ai.com/adaface returns 404.
- Public statementCaptured Sep 25, 2026
States that questions are 'designed keeping the Equal Employment Opportunity Commission (EEOC) guidelines in mind to remove unconscious bias', which is a design claim with no published outcome testing behind it.
- Public statementCaptured Sep 25, 2026
Lists 'EEOC compliant' among its enterprise features without any supporting audit, validation study or adverse-impact analysis.
FRIA Support
2 items30
- AbsenceCaptured Sep 25, 2026
We searched the Trust Guide, DPA, help center and sitemap and found no EU AI Act Article 27 FRIA template, AI Act deployer guidance or instructions for use; no Adaface page mentions the AI Act.
- DocumentationCaptured Sep 25, 2026
Commits to 'reasonable cooperation and assistance' with customers' GDPR Article 35 DPIAs and Article 36 consultations, supplying data categories, retention controls, sub-processor lists, transfer mechanisms and security measures.
Data Governance Disclosure
6 items52
- DocumentationCaptured Sep 25, 2026
The DPA, effective May 2025, names 11 sub-processors with their purposes and locations and says API data sent to OpenAI, Anthropic and Gemini is not used to train models; it also discloses that every sub-processor except DigitalOcean keeps processing outside the EEA, even for EU-hosted customers.
- DocumentationCaptured Sep 25, 2026
Discloses that data is hosted with DigitalOcean in the US by default, that AES-256-encrypted backups keep deleted records for up to 14 days, and that Adaface analyzes 'aggregated and disassociated Customer Data' to improve the product.
- DocumentationCaptured Sep 25, 2026
Offers dedicated deployments in the US, Netherlands, Germany, UK, Canada, Singapore or India on request at extra cost; none is enabled by default on any plan.
https://www.adaface.com/trust/data-management/data-residency
- DocumentationCaptured Sep 25, 2026
States that by default candidate data is stored 'until it's explicitly removed', with periodic deletion set up contract by contract.
- DocumentationCaptured Sep 25, 2026
States that by default Adaface does not collect résumés or candidates' social profiles, so employers are not biased when they analyze screening data.
- AbsenceCaptured Sep 25, 2026
The Trust Guide, DPA and help center show no ISO 27001, SOC 2 or ISO 42001 certification, no statement of which candidate data reaches the LLM scorers or which attributes are excluded, and no privacy notice for the Honestly desktop app.
Human Oversight Design
6 items50
- DocumentationCaptured Sep 25, 2026
Says candidates are 'scored in real-time & automatically given a pass/ border-line pass/ fail status' on a scorecard with a skill breakdown and industry benchmarks.
- DocumentationCaptured Sep 25, 2026
Recruiters can review the score distribution of candidates who have finished a test and adjust its cut-off score in the test settings.
https://help.adaface.com/en/articles/6336349-how-do-i-change-the-cut-off-score
- DocumentationCaptured Sep 25, 2026
Describes proctoring screenshots for recruiters to review and calls the test 'just a first filter' that is 'focused on elimination, not selection', leaving interviews to catch cheaters.
https://help.adaface.com/en/articles/6456153-how-does-adaface-prevent-candidates-from-cheating
- DocumentationCaptured Sep 25, 2026
Super admins can restrict who changes test and proctoring settings and get 'access to complete audit logs'.
- DocumentationCaptured Sep 25, 2026
The MCP server gives AI assistants read-only access by default, while an opt-in read-and-write mode lets them send tests and triage candidates through a shortlist_or_archive tool.
- Public statementCaptured Sep 25, 2026
Tells candidates 'the evaluation is completely automated' and refers questions about results and retakes to the employer, with no re-review or appeal route described.
https://help.adaface.com/en/articles/6336757-faqs-for-candidates
Post-Market Monitoring
5 items38
- DocumentationCaptured Sep 25, 2026
An unlinked Gatus health dashboard, read through its JSON API, runs automated checks on 31 endpoints, including the API, the app, an 'AI Engine' and code runners, but keeps only recent results and shows no incident history or announcements.
- DocumentationCaptured Sep 25, 2026
The SLA, effective January 2024, commits to 99.5% monthly availability with service credits and severity-based response times of 10 to 48 hours.
- DocumentationCaptured Sep 25, 2026
Commits to notifying personal-data breaches within 72 hours of discovery, and to posting notices of general incidents on its blog and social media.
- DocumentationCaptured Sep 25, 2026
Says Adaface checks every month whether ChatGPT can solve its questions and runs web crawlers to find and replace leaked questions; this monitors question integrity rather than fairness or the scoring model.
- AbsenceCaptured Sep 25, 2026
No security.txt, vulnerability-disclosure policy, dedicated security contact, scoring or model changelog, or fairness-monitoring dashboard was found; the Trust Guide lists only the general ada@adaface.com inbox and an individual's address for legal requests.
Customer Documentation
5 items58
- DocumentationCaptured Sep 25, 2026
The Trust Guide links to privacy, security, architecture, compliance, GDPR, data-management, data-residency and data-request pages, plus a downloadable security and compliance whitepaper.
- DocumentationCaptured Sep 25, 2026
A public help center with 49 articles in 11 collections covers scoring, benchmarks, proctoring, accommodations and candidate FAQs, but none of them addresses AI, bias or regulatory compliance.
- DocumentationCaptured Sep 25, 2026
Explains accommodations employers can grant, such as more time per question, untimed tests, text-only tests and switching off webcam or web proctoring; Adaface forwards candidates' requests to the employer.
- DocumentationCaptured Sep 25, 2026
An October 2023 blog guide covers Title VII, the ADA, the ADEA and EEOC principles for pre-employment testing, but not NYC LL 144, the EU AI Act, Illinois or Colorado.
https://www.adaface.com/blog/is-pre-employment-testing-legal/
- AbsenceCaptured Sep 25, 2026
No NYC LL 144, EU AI Act, Illinois HB 3773 or Colorado deployer guidance, and no candidate AI-notice templates, were found on adaface.com or in the help center.
Editorial notes
Company overview
Adaface Pte. Ltd. is a Singapore-headquartered assessment vendor with an office in Bengaluru. It was founded in 2018 by Deepti Chopra and Siddhartha Gunti and is backed by Entrepreneurs First and SAP.iO. Its chatbot, Ada, delivers 40-minute "conversational assessments" for 500+ roles, covering coding, aptitude, Big Five personality and role-specific skills. It scores candidates automatically and labels each one pass, borderline pass or fail on a benchmarked scorecard. Adaface says 1,200+ companies in 80+ countries use it. Its DPA names OpenAI, Anthropic and Google Gemini as sub-processors for "assessment authoring and scoring". Proctoring includes webcam snapshots, screen capture, tab-switch logs and IP tracking. Honestly, a desktop add-on costing $10,000 a year, detects hidden AI cheating tools. An AI résumé parser helps recruiters work out which candidates are "best suited" for a role. Plans cost $180 to $50,000 a year. A detailed Trust Guide covers GDPR, data residency and security, but Adaface publishes no AI documentation, no bias audit and no security certification.
Regulatory exposure
Adaface labels candidates pass, borderline pass or fail automatically. Its own case study says the top 30% go straight to hiring-manager interviews. NYC employers that use these labels, or the AI résumé parser, to decide who advances may be using an AEDT under Local Law 144, particularly where LLM scoring is involved. That would require an annual independent bias audit, a published summary and notice to candidates. Adaface publishes no audit and no LL 144 material, and DCWP has promised tougher enforcement after the December 2025 Comptroller audit. The EU AI Act treats tools that evaluate and filter job applicants as high-risk under Annex III. Provider obligations now apply from 2 December 2027. They will reach the LLM-based scoring that Adaface's DPA discloses, including technical documentation, logging and accuracy duties. We found no emotion-recognition feature: webcam proctoring captures images for recruiters to review. Since 1 January 2026, Illinois employers must give HB 3773 notice when AI informs employment decisions. We found no AI analysis of recorded video interviews, so the AI Video Interview Act looks less likely to apply. California's FEHA rules require four years of automated-decision records. Colorado's SB 26-189 disclosure duties and California's CPPA ADMT rules both begin on 1 January 2027.
Path to a higher score
The fastest gains would come from six steps. (1) Commission an independent bias audit of pass/fail outcomes that covers LLM-scored question types and the résumé parser. Publish impact ratios with the auditor's name and the audit date, and repeat the audit every year. (2) Publish a system card that says which answers each LLM scores, the rubric it uses, how closely it agrees with human raters and its known limits. Add Adaface's own validity and adverse-impact data instead of general Schmidt and Hunter figures. (3) Document human-oversight controls: manual score override, human review of LLM-scored answers, a setting that turns off automatic pass/fail labels, and a way for candidates to request re-review. (4) Add EU AI Act instructions for use, a FRIA template and candidate-notice templates for NYC, Illinois and Colorado. (5) Link the status page and add incident history, then publish a vulnerability-disclosure route and a changelog for scoring and the question bank. (6) Support or remove the "EEOC compliant" badge, and obtain ISO 27001 or SOC 2 certification.
Regulatory frame
What applies to assessment AI.
Assessment is the most-regulated surface in AI hiring. NYC LL 144, EEOC and OFCCP frameworks, Illinois HB 3773, and EU AI Act Annex III §4 all apply, with overlap on bias-audit obligations.
Compare
Adaface against its nearest-scoring peers.
The closest scores in Assessment AI, side by side.
- Side-by-sideAdaface vs Pillar40F
- Side-by-sideAdaface vs Feenyx40F
- Side-by-sideAdaface vs Canditech39F
- Side-by-sideAdaface vs DevSkiller39F
Ranking
Others rated in Assessment AI
Ranked 67 of 82 by weighted total under rubric v1.0. The ordering is arithmetic on the rubric and carries no view on which tool suits a given hiring process.
- #64Pillar40F
- #65Feenyx40F
- #66Cappfinity40F
- #67AdafaceThis profile39F
- #68Canditech39F
- #69DevSkiller39F
Conflicts of interest
No vendor pays for placement, scoring, or removal. Casework - the consulting firm that operates this directory - provides paid services to some vendors. Any active or recent (within 24 months) commercial relationship is disclosed on the affected vendor profile and the review is reassigned to an independent reviewer. See the full policy on About.
Casework has no commercial relationship with this vendor.