Trust
Security
- Every connection is encrypted, and stored data and recordings are encrypted too.
- Each firm's candidates are walled off from every other firm.
- No one enters an interview room without agreeing to be recorded.
- Signing in needs a password and a code sent to your email.
- We back up every night and test restores, and we email you within 72 hours if a breach affects your data.
If your security team has a questionnaire, send it to admin@hireanex.com and we'll answer it in writing. A Data Processing Agreement is available for signature on request.
Last updated: November 17, 2026
What we protect
Hireanex handles candidate data on behalf of hiring teams: contact details, resumes, interview audio and video, transcripts, and evaluation reports. Protecting that data is the product — a hiring platform that leaks candidate interviews has failed at its one job.
Encryption
Every connection uses TLS — the web app, the API, video calls, and recording playback. Data is encrypted at rest in the database (Neon Postgres) and in recording storage (Cloudflare R2). Database backups are additionally encrypted at the application layer before they leave the database provider.
Your data stays yours
- Every request is scoped to your organization — candidate data is partitioned per customer and never visible across organizations.
- Within your organization, role-based permissions apply: by default recruiters can add candidates to a teammate's open job but see contact details and reports only for their own candidates; admins control per-role access.
- Report share links use unguessable tokens, expire after 90 days unless the firm chooses otherwise, and are revoked when a candidate is deleted. A shared report leaves out the screening and logistics summary, and any part of the interview about pay or work authorization. A client can answer on the report; answers are limited per network address and per link, and the page never shows the comment or name back.
- An append-only audit log records sign-ins to your accounts (including failed attempts), admin changes, and every access to candidate data — with who, when, and from where. Entries can't be edited or deleted, and admins can search the log in-app and export it as CSV.
- Admins can export a copy of their organization's data — candidates, reports, transcripts, and resumes, plus download links for interview recordings — as a self-serve download, any time.
Recording with consent
A candidate cannot enter the interview room without explicitly consenting to being recorded — the check is enforced on the server, not just in the interface. The setup checks before an interview record nothing: the video-connection test sends a generated test pattern to a separate test room for a few seconds, never the camera or the microphone. Recordings live in a private bucket with no public access and playback uses short-lived signed URLs. Recordings are retained while your account is active. An admin can delete a candidate's records from our live systems on the candidate's page, and the audit log records who did it and when. That includes the interview's per-turn technical record (what the interviewer did, which provider answered and timings, with no candidate words) and the optional rating and note a candidate leaves about the interview. If you instruct us in writing instead, we delete them from our live systems within 30 days. Backup copies are retained separately and we do not commit to a timetable for removing data from them.
Account security
- Two-factor sign-in for every account on new devices: a password alone is never enough — signing in also requires a 6-digit code sent to the account holder's email. A browser that has passed the code is trusted for 30 days; changing or resetting the password revokes that trust.
- Passwords are hashed with bcrypt; signup and password reset require email verification with lockouts on repeated failures.
- Changing or resetting a password immediately signs out every other session — a stolen token dies with the old password.
- Deactivating a team member takes effect immediately, not when their session expires, and takes them out of any interview that is running.
- On the Enterprise plan, firms can require company sign-in (SSO) through their own identity provider, such as Okta, Microsoft Entra or Google Workspace, for everyone on their verified email domains. One named backup admin keeps a password and emailed code in case the provider fails; every use is logged and emailed to the firm's other admins.
- Firms can connect their directory so people join and leave Hireanex with their staff record. Someone removed there loses access within about a minute; a firm's last admin and its backup admin are never removed this way.
- Sign in with Google or Microsoft only ever signs in someone who already has an account, confirmed once by a code sent to their email.
Backups you don't have to take on faith
The database supports point-in-time recovery. Every night we take an encrypted database dump, stored outside the database provider, and a second copy of all recordings in a separate backup bucket. The restore procedure isn't theoretical — we run restore drills and the last one passed in July 2026.
How we build
Every change runs through continuous integration: type-checked builds, an automated test suite that exercises the real API against real databases, and dependency vulnerability scanning. Production releases are gated on green CI. AI-generated scores are range-validated on the server before they are stored, and model inputs are hardened against prompt injection. When a recruiter starts a job from a posting link, the server reads only public https pages, refuses private and internal addresses at every redirect, and caps each read in time and size. Our email provider tells us about an email that could not be delivered through a signed webhook, which the server reads only when the signature matches.
AI data handling
Speech becomes text with Deepgram. Reports, and the questions Interview Assist suggests to an interviewer during a live call, are written with Anthropic's Claude and Google's Gemini, under their paid business terms. Interviews and applications are never used to train AI models. We have not opted in to any provider data-sharing or model-improvement program. AI output is decision support: the platform never auto-rejects anyone — your team records the verdict.
When an organization selects a different AI model for writing interview questions, those question-writing requests are routed through OpenRouter (an AI model gateway) with zero data retention, and may be served by OpenAI or xAI models. They carry the job and interview-plan details and, for personalised questions, an extract of the resume. Every request refuses hosts that store prompts, and a model with no zero-data-retention host is not used: our default model writes the questions instead. We have not opted in to any data-sharing or model-improvement program with these providers. Evaluation and scoring stay on our core providers regardless of that choice.
In AI-led interviews, Anthropic's Claude writes the interviewer's replies. It receives the candidate's speech as text, the job description, the interview plan and a resume summary. If Anthropic has not started to answer within 3 seconds and our backup route is switched on, the same Claude model answers from Amazon Bedrock or Google Vertex AI in the United States, reached through OpenRouter with zero data retention (nothing is kept after the answer). If that also fails, Google Gemini answers. The backup route is off unless we switch it on. We log which provider answered; the report never shows it.
The in-app help chat for recruiters and admins runs on Anthropic's Claude. It answers from our help articles and has no access to candidate data.
The hiring assistant
Firms that use the hiring assistant post one link per job. Candidates confirm their email with a 6-digit code, then answer the firm's screening questions in a chat, add their resume, ask about the job and book their interview. The first screen and the chat's first line say it is an AI and that the hiring team decides. The chat, the answers, any questions for the recruiter and the resume are stored with the candidate's application in the firm's account, under the same isolation, permissions, export and deletion as the rest of its candidate data. A number typed in the shape of an Aadhaar, PAN or Social Security number in a candidate's messages is hidden before it is stored.
Most of the chat is scripted. When the chat cannot read something a candidate typed by itself, Anthropic Claude reads that message and the job's own facts (with the end client's name removed, and pay only if the firm shows it) to tell whether it is an answer or a question, and to draft a reply to a question about the job. The candidate's name, contact details and resume are not sent with it, though Claude sees whatever the message says. Every reply is checked before the candidate sees it; anything the job's facts do not answer goes to the recruiter instead.
The assistant never turns anyone away. Screening answers are compared with the answers the firm requires by fixed rules, not by an AI model. When an answer does not meet a requirement, the candidate is told which one, can ask the recruiter to review, and nothing is booked until a person decides. The recruiter sees how many answers match, never a score or a ranking, and the answers never change the interview score. A candidate's answers about pay and work authorization, in the chat or to the interviewer's questions, are left off the written report and transcript shared with the firm's client. The recording, if the client can play it, is not edited.
The chat and its answers are kept while the firm's account is active, including for applicants who are never interviewed. An application left unfinished for 7 days is closed, and its data stays with the firm. An admin can delete a candidate's application from our live systems. Audit-log entries remain, and copies in our nightly database dumps and backup bucket are retained separately, with no committed timetable for removing them.
Fairness and adverse impact
The scoring model is never given the candidate's name as an input field. This is enforced by an automated counterfactual test: it composes the exact model input for two candidates who differ only in name and fails the build unless the two are byte-identical. (A name the candidate speaks aloud in the interview, or writes into their own resume summary, is part of the content being evaluated and is not removed.)
Hireanex does not collect demographic data about candidates — race, ethnicity, sex, age, or disability status — so we cannot measure adverse impact ourselves, and we say so plainly rather than imply otherwise. Selection decisions and any adverse-impact analysis belong to the employer. To support that analysis, the self-serve data export includes a per-session scores file (scores.csv) with each score, its basis, and the recorded recruiter decision — designed to be joined on candidate email with demographic data you hold, for example voluntary self-identification collected in your ATS.
Who has access internally
CovetIT Inc is a small company, and we say so plainly: production access is limited to two engineers on our staff. A tester can view the status of our web hosting, with no access to the database, stored files or any credential. No contractor or third-party staff has production access, and the AI coding tools our engineers use act only under that engineer’s own account.
Our code organization requires multi-factor authentication for every member, and every infrastructure account has it turned on. Only our founder can release a change to production.
If something goes wrong
All services report to centralized error and latency monitoring with alerting. If we become aware of a breach affecting your data, we notify affected account holders by email without undue delay — and no later than 72 hours after we become aware. This commitment is in our Terms and in the DPA.
Compliance status, honestly
A GDPR/CCPA Data Processing Agreement is available for signature, and our privacy policy keeps the full list of providers (subprocessors) under "Who can see your data" — account holders are emailed before that list changes. The subprocessors page shows, for each provider, what it does, the data it handles and where it runs. We are not yet SOC 2 or ISO 27001 certified; that is planned as the company grows. In July 2026 we completed a 20-track internal security audit and remediated its critical findings.
Reporting a security issue
Email security@hireanex.com with what you found, where, and the steps to reproduce it. We acknowledge within 2 business days and will keep you informed through remediation. This page is our disclosure policy, and the same contact is published in our security.txt.
- Please don't test against production candidate data, and stop as soon as you have shown the problem: don't open, change, download or keep other people's data.
- Please don't run tests that could slow or interrupt the service for other people, such as denial-of-service or high-volume scanning.
- Report it to us first, and give us a reasonable time to fix it before you share the details publicly.
- Problems in a provider's own systems are best reported to that provider. We don't run a paid bounty program.
If you act in good faith and follow this policy, we will not take legal action against you for your research, and we will work with you to understand and fix the issue.