Workstream

The public VoiceAI admin article (updated 14 July 2026) is the closest thing to instructions for use: the AI compares each answer with the employer's 'expected response' and marks it Matched or Unmatched. The article reports the share matched as the score (3 of 4 = 75%), shows 'Low confidence' or 'Insufficient evidence' flags, and lists English, Spanish, Mandarin and French support.

https://help.workstream.us/en/articles/12663091-set-up-and-use-voiceai

Workstream's public VoiceAI/VideoAI interactive demo shows the VideoAI results panel stating that the 'Score is calculated equally across all questions, based on the ratio of passed questions to total questions', and that only questions with expected responses are scored.

https://www.workstream.us/explore-voiceai-videoai

The AI Functionality Terms (last updated 21 June 2024) cover generative AI, LLM and other ML features and disclose that some run on third-party models, pointing to Microsoft Azure's terms. They warn that outputs are 'not tested, verified, endorsed or guaranteed to be accurate', which makes them a liability document rather than technical documentation.

https://www.workstream.us/aiterms

No model or system card, AI principles or responsible-AI page, explainability statement, accuracy or limitation metrics, or ISO/IEC 42001 certification was found. Places checked: the 2,141-URL workstream.us sitemap (/responsible-ai and /ai-principles return 404), the help center, the security page and the legal terms.

Have a URL that disputes this?

No NYC Local Law 144 bias-audit summary, independent audit (Warden AI, Holistic AI, BABL AI, DCI, ORCAA, Credo AI, ConductorAI) or adverse-impact/selection-rate disclosure was found for VoiceAI, VideoAI or resume match scores. trust.warden-ai.com/workstream returns 404, help-center searches for 'bias', 'discrimination', 'EEO' and 'Illinois' return no articles, and the product, pricing, security, franchise and legal pages say nothing on the subject.

Have a URL that disputes this?

The AI terms admit outputs 'may contain biases' and bar customers from discriminatory uses, but they make checking outputs solely the customer's job and disclose no fairness testing by Workstream.

https://www.workstream.us/aiterms

Workstream markets VoiceAI as knowing 'what questions it can and can't ask applicants' and so keeping employers 'compliant with hiring laws', an unaudited claim about question content rather than outcome testing.

https://www.workstream.us/artificialintelligence

The VoiceAI product page quotes a customer who credits automated phone screening with 'filtering out' lukewarm candidates, which shows the tool works as a hiring gate and not just a scheduler.

https://www.workstream.us/product/hiring/voiceai

Workstream's own June 2026 buyer's guide credits competitor HeyMilo with 'Third-party bias auditing through Warden AI' but lists no audit or fairness testing for Workstream.

https://www.workstream.us/blog/conversational-ai-recruiting-software-qsr

A 2023 Workstream blog post claims automated systems can 'reduce and eliminate bias by treating every candidate equally', citing an unlinked University of Chicago study rather than any test of Workstream's own tools.

https://www.workstream.us/blog/automated-hiring

No EU AI Act reference, Article 27 FRIA template, deployer guide or AI-law compliance kit was found on workstream.us, the help center or the franchise page. The legal documents checked (Subscription Services Agreement of 23 September 2026, AI Functionality Terms, privacy policy of 5 May 2026) contain none either.

Have a URL that disputes this?

Section 7 of the AI terms bars customers from using AI for 'automated decision-making that has legal or similarly significant effects' without adequate human review, and from exploiting group vulnerabilities based on age or physical or mental characteristics. The wording echoes EU AI Act Article 5 without naming the Act.

https://www.workstream.us/aiterms

The Subscription Services Agreement makes the customer responsible for complying with Title VII, the ADEA, the ADA and any applicable employment, data-protection or privacy law, which hands deployer duties to the customer without supporting materials.

https://www.workstream.us/terms

The security page claims SOC 2 Type II with annual independent audits, AES-256 encryption at rest and TLS 1.2+ in transit, AWS hosting mainly in us-west-2 with multi-AZ backups, tenant isolation, and audit logs of 'every read, write, and administrative action'. It also says Workstream does not train shared AI models on customer data without explicit opt-in approval.

https://www.workstream.us/product/platform/security

The AI terms bar training on AI inputs and outputs except for the customer's benefit, but they allow de-identified inputs and outputs to be used to train AI 'or for any other lawful purpose'. They also let third-party model providers use the data as needed to run the features.

https://www.workstream.us/aiterms

The privacy policy (updated 5 May 2026) discloses sharing with 'AI technology providers and voice processing services' for AI phone tools but excludes data processed for employer customers. Its body says nothing about how long call or video recordings are kept, voiceprints or other biometric identifiers, BIPA, or automated decisions about applicants.

https://www.workstream.us/privacy

Section 5.6 of the Subscription Services Agreement commits Workstream to processing Customer Data only on the customer's instructions, but no standalone DPA or subprocessor list is published.

https://www.workstream.us/terms

No data-exclusion list (for example, protected attributes kept out of match scoring), no description of VoiceAI or VideoAI training data, no subprocessor list or DPA (workstream.us/subprocessors and /dpa return 404), and no public trust center were found.

Have a URL that disputes this?

For each question, reviewers see the applicant's answer, a pass/fail against the employer's expected response, the full transcript and call recording, 'Low confidence' or 'Insufficient evidence' warnings, and AI notes labelled 'Chatbot'. Admins can hide the matching score, and auto-advance on completion is off by default.

https://help.workstream.us/en/articles/12663091-set-up-and-use-voiceai

Workstream's public demo shows the VideoAI results panel listing each question with its expected response, a Pass/Fail verdict, an AI-written rationale and a 'View full conversation' link, with the note 'We recommend to look at each question closely.'

https://www.workstream.us/explore-voiceai-videoai

Stage permissions (updated 27 August 2026) can stop a stage such as VoiceAI from being skipped and limit who may move applicants in or out, for example to enforce 'manager sign-off'. However, 'Automations and Super Admin accounts are not affected by these restrictions'.

https://help.workstream.us/en/articles/16712514-set-up-hiring-stage-permission-controls

Workstream's Mira AI assistant cannot message or reject applicants from unattended scheduled tasks; such actions are excluded 'for safety' and must be approved in a live chat.

https://help.workstream.us/en/articles/16418410-set-up-scheduled-tasks-in-ask-ai

Managers can record rejection reasons and restore ('Unreject') rejected applicants, which gives a way to reverse rejections informed by the AI.

https://help.workstream.us/en/articles/11626020-applicant-outcomes

The Application Form stage can 'automatically reject' applicants who don't meet the employer's screening criteria as soon as they submit, so fully automated rejection is still available alongside the AI stages.

https://help.workstream.us/en/articles/12633337-set-up-the-application-form-stage

The Applicants page shows a resume 'match score' for how well each resume fits the job, with no explanation of how it is calculated or how to challenge it.

https://help.workstream.us/en/articles/12513649-spot-new-applicants-and-review-top-matches

Workstream's public Atlassian Statuspage, linked from its help-center footer, lists 19 components with email, Slack and RSS subscriptions but no VoiceAI or VideoAI component. The components include Hiring, Career Page, Onboarding, Payroll, Twilio SMS/MMS, SendGrid and the ADP, UKG, Paylocity and QuickBooks integrations.

https://workstream-status.statuspage.io/

The incident feed lists 12 incidents from April 2022 to 5 July 2024, two of them with postmortems (a June 2023 jobs-site outage and December 2023 onboarding latency), and none in the roughly 26 months since.

https://workstream-status.statuspage.io/api/v2/incidents.json

No security.txt (workstream.us/.well-known/security.txt returns 404), vulnerability-disclosure policy, bug bounty or dedicated security contact was found; the only contacts are privacy@workstream.us and help@workstream.is. There is also no AI model-update changelog and no fairness or drift monitoring.

Have a URL that disputes this?

The security page says Workstream regularly scans for vulnerabilities and logs and reviews access to production systems, but these are internal controls with no public reporting.

https://www.workstream.us/product/platform/security

The public VoiceAI/VideoAI product page describes AI phone screens with 'pass/fail ratings' and recorded calls, transcripts and AI summaries. It also describes VideoAI's video replay, conversation summary and 'Candidate match score'.

https://www.workstream.us/product/hiring/voiceai

The public Intercom help center has 173 product articles and 48 guides, including VoiceAI setup and scoring and hiring-stage automations and permissions, plus links to the API docs and the status page.

https://help.workstream.us/en/

The quote-based Hiring plan lists 'VoiceAI screening', 'VideoAI screening', 'Automate applicant tracking and scoring', and 'Call transcripts, recordings, and AI summaries and match scores'.

https://www.workstream.us/pricing

The franchise page documents role- and entity-based permissions and franchisee-franchisor visibility controls for multi-location operators, but gives no AI compliance guidance.

https://www.workstream.us/franchise

None of the following was found: an NYC LL 144 candidate-notice template, Illinois AI Video Interview Act notice, consent or deletion guidance, BIPA guidance for recorded calls and videos, Colorado or EU AI Act deployer guidance, a VideoAI help article, a public DPA or a subprocessor list.

Have a URL that disputes this?