Hey Juno

How do you assess the security of an AI research platform?

By Hey Juno · Published · Updated

To assess the security of an AI research platform, ask for its independent audit report, its data processing agreement, and written answers on access control, retention, subprocessors, AI model providers and whether your data trains models. Then check that each answer is documented, not just promised. A certification is a good start, but it is not the whole review.

Customer interviews hold names, opinions, recordings and sometimes sensitive detail. The vendor processes that data on your behalf, so its weak points become yours. This guide gives a practical order of work for teams without a security specialist. For the data itself, see PII in customer research.

Start with what a certification does and does not prove

SOC 2 reports come from an independent auditor and sit within the American Institute of CPAs (AICPA) family of SOC services. In general, a Type I report looks at whether controls are designed properly at one date. A Type II report also tests whether they worked over a period of time, which is why it carries more weight.

What it tells you:

  • An outside auditor tested controls against criteria such as security and confidentiality.

  • The controls operated across the audit period, not just on audit day.

What it does not tell you:

  • That every control you care about is in scope. Read the scope and the list of trust criteria covered.

  • That the product is free of vulnerabilities or has never had an incident.

  • That the vendor's AI model providers and other suppliers are covered. Reports often treat them as separate.

  • How your own settings should be configured. That remains your job.

Ask for the report itself, not a badge. Read the auditor's opinion, the system description and any exceptions noted.

Ask for the data processing agreement

If you handle personal data of people in the EU or UK, privacy law generally expects a written contract with any processor acting for you. The GDPR sets out what that contract must cover (Article 28). Other laws, such as Australia's Privacy Act, set their own rules (OAIC). A data processing agreement (DPA) usually sets out:

  • what data is processed and for what purpose

  • security measures the vendor must keep

  • whether and how subprocessors are used

  • what happens at the end of the contract, such as return or deletion

  • how breaches are notified to you

Check it against your own legal or privacy team's requirements before you collect any data.

Check identity and access control

Most real-world data exposure comes from access, not from a clever attack. Ask:

  • Single sign-on (SSO). Can your team sign in through your own identity provider, so offboarding someone removes their access everywhere?

  • Multi-factor authentication (MFA). Is it available, and can it be required?

  • Roles. Can you separate who creates studies, who sees raw responses and who exports data?

  • Approvals. Can sensitive actions, such as exports, be routed for sign-off?

  • Sharing. How do shared links and external collaborators work, and can access be revoked?

Look at retention and deletion

Ask how long the platform keeps interview data, who sets that period, and how deletion works. Ask whether deletion reaches backups and exports, and how long that takes. A good answer is specific.

Ask where the data lives

Data residency is where your data is stored and processed. Some contracts and laws care about it, and some do not. Ask which regions the vendor uses, whether you can choose, and whether support staff or AI processing happen elsewhere. Get the answer in writing, because location can differ between storage, processing and backup.

Ask about subprocessors and AI model providers

An AI research platform usually relies on other companies: cloud hosting, speech and transcription services, and AI models. Each one may touch your participants' words. Ask for:

  • the current list of subprocessors and what each does

  • how you are told when the list changes

  • which AI model providers see interview content

  • whether those providers keep or reuse the content

Ask directly whether your data trains models

This is the question people most often skip, and it deserves a clear written answer. Ask separately about:

  • the vendor training its own models on your interviews

  • third-party model providers training on content sent to them

  • any use of your data to improve the product, even if it is not called training

Participant consent matters here too. If you told participants their words would be used only for your study, a vendor that reuses them creates a problem for you.

Check the audit trail

An audit trail records who viewed, changed or exported what, and when. Ask what is logged, how long logs are kept and whether you can see them. It matters most after an incident and for showing a client that access was controlled.

How Hey Juno answers these questions

Hey Juno is the company behind Juno, an AI researcher that runs qualitative interviews. These are the facts Hey Juno publishes. Anything not listed here, ask Hey Juno directly.

  • Compliance. Hey Juno lists SOC 2 Type II, GDPR and a DPA among its security measures. Its Trust Centre at trust.heyjuno.co holds control status, audit reports and document requests. Confirm what is current there, and request the documents your review needs.

  • Access. Single sign-on and multi-factor authentication are available. Roles and approvals decide who can create studies, view responses and export data, and sensitive actions can be routed through approvals. Teams work inside one organisation.

  • Data handling. Participant data is encrypted in transit and at rest. You set how long answers are kept and can have them deleted.

  • PII redaction. Juno has built text redaction of hard identifiers such as names, emails and phone numbers, but it is a platform-wide setting that is not switched on in production today. Transcripts, quotes and exports are not redacted, and customers cannot switch it on themselves. Switching it on is planned, and so is redacting audio. Neither has a date. Until then, treat transcripts and exports as containing personal information. Access is limited to people in the customer's organisation who can see the study, and participants agree to recording before a voice interview starts. The details are in PII in customer research.

  • Participant data in the interview. Juno captures what the study needs and no more, and handles answers about health, finances and family with tighter care. See the safety standards.

Hey Juno's public pages mention regional data options. They do not name regions, subprocessors, AI model providers or a policy on model training. Ask for those in writing as part of your review, using the questions above.

A procurement checklist

  • Request the latest SOC 2 Type II report and read the scope and exceptions.

  • Get the DPA and have your legal or privacy team review it.

  • Confirm SSO and MFA, and test offboarding.

  • Map roles to your team's real jobs, including who can export.

  • Get retention periods and deletion steps in writing.

  • Get data regions for storage, processing and backups.

  • Get the subprocessor and AI model provider lists, and the notification process.

  • Get a written statement on model training, for both the vendor and its providers.

  • Ask what the audit trail records and how you can see it.

  • Ask how breaches are reported to you, and how fast.

  • Check that the vendor's participant-facing wording matches what you promise participants.

Where this doesn't apply

This is a starting checklist, not a security audit or legal advice. Regulated sectors, government work and research with children or health data may require more, such as a formal vendor assessment or penetration test results. Check with your own legal, privacy and security teams. The guide also cannot tell you whether a particular vendor is safe. Only the documents and your own review can.

Frequently asked questions

Does SOC 2 Type II mean a platform is secure?

It means an independent auditor tested the in-scope controls over a period. It does not cover everything, and it does not replace reading the scope, the exceptions and your own checks.

What is the difference between SOC 2 Type I and Type II?

Type I checks whether controls are designed properly at one point in time. Type II also tests whether they operated over a period.

Do I need a data processing agreement for customer interviews?

If the vendor processes personal data for you, privacy laws such as the GDPR generally expect a written agreement. Confirm what applies to you with your own legal or privacy team.

What should I ask about AI models and my data?

Ask which model providers see interview content, whether they keep it, and whether anyone trains on it. Get the answer in writing.

Where can I check Hey Juno's current compliance status?

On the Trust Centre at trust.heyjuno.co. It holds control status, audit reports and document requests. Hey Juno does not offer a downloadable DPA on its public pages, so ask for it there.