Braintrust AIR
HireAIScore rates Braintrust AIR 59 out of 100 (grade F, Substantial gaps) under rubric v1.0, last reviewed September 25, 2026. Braintrust AIR ranks 8 of 160 vendors rated in Screening AI, against a category average of 46. That score is drawn from 32 evidence items across 7 rubric criteria for Braintrust AIR — 29 with a cited source, 3 recording that nothing was located.
Score breakdown
Braintrust AIR, 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 49
58
+11.6
Bias Audit Transparency
Weight 18% · category avg 41
75
+13.5
FRIA Support
Weight 15% · category avg 31
35
+5.3
Data Governance Disclosure
Weight 15% · category avg 53
56
+8.4
Human Oversight Design
Weight 12% · category avg 56
62
+7.4
Post-Market Monitoring
Weight 12% · category avg 40
62
+7.4
Customer Documentation
Weight 8% · category avg 58
63
+5.0
Total weighted score
58.7(rounded to 59 for display)
Category avg is the mean raw score on that criterion across the 160 Screening AI vendors in scope of this rubric, this one included.
Strongest criterion
Bias Audit Transparency
75raw / 100
Contributes 13.5 to the total. 75 against a 41 category average.
Weakest criterion
FRIA Support
35raw / 100
Contributes 5.3 to the total. 35 against a 31 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
- 32
- Documentation
- 18
- Audit report
- 4
- Public statement
- 7
- Absence
- 3
- With a source URL
- 29 of 32
- Source hostnames
- 6
- Fewest items
- 4
- Article 11 Technical Documentation
These figures measure how thoroughly Braintrust AIR was reviewed, not how Braintrust AIR performed — an absence row, recording that nothing was located, is counted like any other item.
Article 11 Technical Documentation
4 items58
- DocumentationCaptured Sep 25, 2026
Braintrust's Responsible AI Policy (last updated 1 May 2026) frames AIR as decision support, lists governance processes such as risk assessments, bias testing and change management, cites the NIST AI RMF, ISO/IEC AI principles and the EU AI Act, and says EU AI Act readiness work was targeted for completion by 1 August 2026.
https://www.usebraintrust.com/air-artificial-intelligence-policy
- DocumentationCaptured Sep 25, 2026
A public help-centre 'EU AI Act Technical File Template' (updated 27 Jan 2026) maps AIR to Arts. 9-15 and 43, including LLM-based scoring of transcripts, audio and video processed for transcription, audit logs kept for at least six years in AWS US West, and a quarterly test of AI scores against recruiter ratings; however, it is a fill-in template for clients with stale entries such as a SOC 2 audit marked 'IN PROGRES'.
https://support.usebraintrust.com/api/v2/help_center/en-us/articles/33456665169815.json
- DocumentationCaptured Sep 25, 2026
Help-centre guidance tells deployers what AIR cannot do, including that it 'does not verify candidate claims or credentials' and scores only content against recruiter-defined criteria.
https://support.usebraintrust.com/api/v2/help_center/en-us/articles/33453886785943.json
- AbsenceCaptured Sep 25, 2026
The AIR product, compliance, policy, release-notes, help-centre and Trust Center pages contain no model card completed by Braintrust, no named foundation model or version, no validity study linking AIR scores to job performance, and no ISO/IEC 42001 certification.
Bias Audit Transparency
5 items75
- Audit reportCaptured Sep 25, 2026
Warden AI's public NYC LL 144 summary for Braintrust's AI Interviewer (report date 15 Aug 2026) reports sex, race/ethnicity and intersectional impact ratios of 0.97-1.00 on 5,040 test records from Warden's own dataset of real candidate profiles, used 'due to a lack of access to historical data'.
- Audit reportCaptured Sep 25, 2026
Warden's AI Interviewer dashboard claims monthly auditing and shows 58 'Clear' results on a 28,980-record test set covering sex, race/ethnicity, intersectional, age, disability, religion and sexual orientation, but only one evaluation (15 Aug 2026) is visible and veteran status and national origin are not tested.
- Audit reportCaptured Sep 25, 2026
Holistic AI's full NYC LL 144 bias audit of Braintrust AIR, dated 23 Sept 2025, is also viewable without a request in the vendor's Trust Center; it found impact ratios of 0.98-1.00 across gender, ethnicity, age, disability and veteran status, using 20,000 synthetic transcripts made by adding demographic self-introductions to real interview templates.
https://drive.google.com/file/d/1hk5QdxnecwI94ppp7HvxOcTZNEZBPuJs/view
- Public statementCaptured Sep 25, 2026
The AIR compliance page headlines 'Zero bias detected' and 'EEOC Compliant' and lists veteran status as 'No Exceptions', while its 'View Full Audit Report' button opens Warden's resume-screener dashboard, which does not test veteran status.
- Public statementCaptured Sep 25, 2026
A June 2025 blog post describes an earlier audit of 'historical interview evaluations' by an unnamed 'independent firm' whose report goes only to enterprise clients on request, so it is not credited.
https://www.usebraintrust.com/blog/independent-ai-bias-compliance-audit
FRIA Support
4 items35
- DocumentationCaptured Sep 25, 2026
The help-centre EU AI Act Technical File Template gives deployers a structured checklist for Arts. 9-15, but it is not a fundamental-rights impact assessment and it places provider duties (technical file, EU database registration) on the client.
https://support.usebraintrust.com/api/v2/help_center/en-us/articles/33456665169815.json
- DocumentationCaptured Sep 25, 2026
The Responsible AI Policy lists customer responsibilities (human review of outputs, candidate notices, accommodations or alternative processes, legal compliance) but gives no guidance specific to Art. 26 or Art. 27.
https://www.usebraintrust.com/air-artificial-intelligence-policy
- Public statementCaptured Sep 25, 2026
The AI Regulations page says clients receive 'DPIA/AIRA' impact-assessment kits and notice templates through the Trust Center, but none appear among the Trust Center's listed resources.
https://www.usebraintrust.com/how-braintrust-air-stays-compliant
- AbsenceCaptured Sep 25, 2026
No Article 27 FRIA template or deployer guide appears on the compliance, AI regulations, policy, help-centre or Trust Center pages (Trust Center resources reviewed in a rendered browser on 25 Sept 2026).
Data Governance Disclosure
5 items56
- DocumentationCaptured Sep 25, 2026
The AIR Privacy Policy (updated 21 Sept 2026) says AIR evaluates interviews with licensed pre-trained foundation models and never uses recordings, transcripts or results to train any model; it performs no facial recognition, voice-print matching or emotion inference, hosts data in the US or EU under SCCs, and keeps records for the engagement plus one year.
- DocumentationCaptured Sep 25, 2026
The Vanta Trust Center shows SOC 2 Type II (report on request), controls for data retention, classification and deletion, and five sub-processors (AWS, Retell AI, Kombo, WorkOS, LiveKit), none of which is named as the provider of the scoring foundation model.
https://app.vanta.com/usebraintrust.com/trust/b7n730pooch9160wkcu31g
- Public statementCaptured Sep 25, 2026
A 2024 AIR FAQ, still published in the help centre, says the models were built in-house and trained on client data and historical hiring records, which contradicts the current statement that no model is trained on candidate data.
https://support.usebraintrust.com/api/v2/help_center/en-us/articles/25854800416663.json
- Public statementCaptured Sep 25, 2026
A help-centre article says AIR's 'scoring model is retrained' when its six-monthly bias audits find discrepancies, which also conflicts with the privacy policy.
https://support.usebraintrust.com/api/v2/help_center/en-us/articles/33453886785943.json
- Public statementCaptured Sep 25, 2026
The AI Regulations page claims data minimisation to 'name + email only', although the privacy policy lists transcripts, video recordings and assessment results, and AIR also imports resumes.
https://www.usebraintrust.com/how-braintrust-air-stays-compliant
Human Oversight Design
5 items62
- DocumentationCaptured Sep 25, 2026
The AIR product page says ranked scorecards and evidence go to the ATS for human review and that 'AIR does not auto-reject candidates'.
- DocumentationCaptured Sep 25, 2026
Release notes document concrete controls: recruiter-editable scoring guides (16 Oct 2025), an optional lock on job configuration after the first interview (2 Jul 2026), a field-level audit trail of job edits (8 Jul 2026) and an activity timeline naming who took each action (15 Jul 2026).
- DocumentationCaptured Sep 25, 2026
The AI Assistance Detection flag shows recruiters a confidence score and a high-level explanation, and the vendor says it never disqualifies, blocks or notifies candidates automatically.
https://support.usebraintrust.com/api/v2/help_center/en-us/articles/37973366765079.json
- DocumentationCaptured Sep 25, 2026
The vendor's recommended invite email tells candidates they can opt out, stop the AI interview at any time and ask for a human-led interview without penalty; the privacy policy adds a right to request human review of any outcome or contest it.
https://support.usebraintrust.com/api/v2/help_center/en-us/articles/33455377115287.json
- Public statementCaptured Sep 25, 2026
A help-centre article tells recruiters that AIR lets them 'Filter out unqualified candidates automatically' and review only top-ranked applicants, which undercuts the claim that AIR never auto-rejects.
https://support.usebraintrust.com/api/v2/help_center/en-us/articles/33848881837591.json
Post-Market Monitoring
4 items62
- DocumentationCaptured Sep 25, 2026
A public Statuspage, 'BT AIR Production', shows live status and 90-day uptime for four components, including the AIR LLM API.
- DocumentationCaptured Sep 25, 2026
The status page's incident history includes a September 2026 AWS outage, a September 2025 API error incident and a December 2025 'Malicious Redirect' security incident traced to a Next.js vulnerability, for which lessons learned were published.
- DocumentationCaptured Sep 25, 2026
Public release notes, dated roughly weekly since 28 May 2025, log scoring-related changes such as an AI engine upgrade (Aug 2025) and enhanced scoring guides (Oct 2025), but give no model or version identifiers.
- Audit reportCaptured Sep 25, 2026
Warden's dashboard advertises continuous monthly bias auditing and 'post-market monitoring' of the AI Interviewer, though only the 15 Aug 2026 evaluation is shown.
Customer Documentation
5 items63
- DocumentationCaptured Sep 25, 2026
The 'How Braintrust AIR Stays Compliant' page maps AIR against FTC guidance, Illinois' AI Video Interview Act, California's rules on automated-decision systems, NYC LL 144, the EU AI Act and GDPR, but carries stale references such as a Colorado 'SB 169 (Effective Feb 2026)'.
https://www.usebraintrust.com/how-braintrust-air-stays-compliant
- DocumentationCaptured Sep 25, 2026
A 29-article AIR help centre covers interview set-up, scoring best practices, review workflows, AI-assistance detection, accessibility disclaimers, invite templates, and guidance on ethics and failure modes.
- DocumentationCaptured Sep 25, 2026
An AIR compliance hub links the Warden bias dashboard and the Vanta Trust Center and summarises SOC 2 Type II, data residency, human review and upfront candidate disclosure.
- DocumentationCaptured Sep 25, 2026
The AIR Terms (last updated 28 Feb 2025) state that AI scores are 'recommendations only' but link no DPA.
- AbsenceCaptured Sep 25, 2026
No public DPA was found, and no candidate-notice template specific to NYC LL 144 (covering the 10-business-day notice, the job qualifications and characteristics assessed, and data-retention information); the vendor says it supplies notice templates to clients.
Editorial notes
Company overview
Braintrust AIR is the enterprise AI-interviewing product of Freelance Labs Inc. (dba Braintrust), a US company founded in 2018. Braintrust launched AIR as a standalone product in 2024; its freelance talent network is not scored here. AIR runs adaptive voice interviews that are recorded and transcribed. It scores the transcribed answers on a 1–5 scale against rubric criteria that recruiters can edit, adds resume-match and AI-assistance signals, and syncs ranked scorecards, transcripts and recordings to more than 50 applicant tracking systems. The vendor says AIR never auto-rejects a candidate and that hiring teams make every decision. Pricing is volume-based, with no public rate card.
Regulatory exposure
Because AIR scores and ranks job applicants, it is a high-risk Annex III point 4 system under the EU AI Act, with obligations now due on 2 December 2027. It is also likely an AEDT under NYC Local Law 144 wherever employers rely on its rankings, and DCWP has promised tougher enforcement after the December 2025 Comptroller audit. Because interviews are recorded, Illinois' AI Video Interview Act and the notice and anti-discrimination duties of HB 3773 apply. California's FEHA rules on automated-decision systems add four-year record-keeping, and Colorado's SB 26-189 adds disclosure duties from 1 January 2027. The vendor excludes emotion inference, facial recognition and voice-print matching, which keeps AIR clear of the Art. 5(1)(f) ban. However, no published test covers how its language-proficiency, communication and AI-assistance signals affect candidates by national origin or speech-related disability. Its compliance page also still cites a Colorado 'SB 169' effective February 2026.
Path to a higher score
The biggest gain would come from a model card and technical file completed by Braintrust itself, naming the foundation model and version, validating scores against job outcomes and listing known limitations. This should replace the fill-in template that shifts provider duties onto customers. Bias testing should add deployment data, cover national origin, English proficiency and speech-related disability, include the AI-assistance flag, and show Warden's monthly history. Braintrust should also reconcile its conflicting statements on training data and name the sub-processor behind the scoring model. It should publish its DPA, NYC notice templates and an Art. 26/27 FRIA kit; the DPIA/AIRA kits it advertises are not in its Trust Center. Finally, it should drop claims such as 'Zero bias detected' and 'EEOC Compliant'.
Regulatory frame
What applies to screening AI.
Treated as high-risk under Annex III §4 when the screening output is decisive. EU AI Act Article 13 transparency obligations apply; deployers must give candidates a way to contest.
Compare
Braintrust AIR against its nearest-scoring peers.
The closest scores in Screening AI, side by side.
- Side-by-sideBraintrust AIR vs HiPeople61D
- Side-by-sideBraintrust AIR vs Sense58F
- Side-by-sideBraintrust AIR vs Tezi58F
- Side-by-sideBraintrust AIR vs Findem58F
Ranking
Others rated in Screening AI
Ranked 8 of 160 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.
- #5HiPeople61D
- #6Alva Labs61D
- #7HeyMilo61D
- #8Braintrust AIRThis profile59F
- #9Sense58F
- #10Tezi58F
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.