In short
- A trust center answers a buyer's security questions before they send them. Large vendors publish certifications, gated audit reports, a subprocessor list and a feed of updates.
- Your host's certifications are not yours. Supabase's security page states the split itself: Supabase handles the infrastructure, and customers are responsible for their applications, including RLS policies, API keys and access controls.
- Without SOC 2 or ISO 27001 you can still publish six things that are true: what was tested, when, by what, what is open, where data lives and who to contact.
- A date makes a claim checkable. "Tested on 4 November 2026, no serious finding open" can be verified. "We take security seriously" cannot.
- Never borrow a badge, never write "compliant" for a standard nobody audited you against, and never present an automated test as a manual pentest.
On this page
What is a trust center?
A trust center, also called a trust page or security page, is the public place where a company shows how it protects the data customers give it. It exists because buyers ask. Before signing, a customer's IT or procurement team wants to know where data is stored, who can reach it, who audited the vendor and what happens in an incident. A trust center answers most of that before the first email, and it shortens the security questionnaire that follows.
The format has settled over the past years. A typical trust center has a public layer anyone can read, and a gated layer where a buyer requests the audit report, usually under a confidentiality agreement. Several vendors sell software to build one. You do not need that software to start. You need true statements with dates.
What well-known trust centers publish
We opened four public trust centers on 13 November 2026 and noted what each one shows. They are large companies with audit budgets. Read them for the structure, since that is what your buyers have learned to expect.
| Company | What its public page shows |
|---|---|
| OpenAI (trust.openai.com) | A list of certifications and frameworks, including SOC 2 Type 2, SOC 3 and several ISO standards. A document library with public and private filters and a "Get access" button: the SOC 2 report is behind a request. Self-assessment questionnaires such as the CAIQ. Sections of controls by theme, a list of subprocessors, a feed of updates and a bug bounty programme |
| Vercel (security.vercel.com) | The SOC 2 Type 2 report with its auditor and the exact period it covers (1 July 2025 to 30 June 2026), SOC 3, ISO 27001, PCI DSS and others. Documents with public and private filters. Named subprocessors, with changes announced. Controls grouped by category. Security advisories and compliance news, with a way to subscribe |
| Supabase (supabase.com/security) | SOC 2 Type 2, HIPAA and ISO 27001 statements, with the reports available to Team and Enterprise customers in the dashboard. Encryption, backups and regions. A plain statement of shared responsibility: the customer is responsible for its own application |
| Atlassian (atlassian.com/trust) | Five areas: security, compliance, privacy, reliability and AI. A separate trust portal holds the documents for vendor risk reviews, and a cloud roadmap lists coming changes |
Four patterns repeat. Evidence comes from a third party: an auditor, a testing firm, a bug bounty. Evidence has a date: Vercel names the period its report covers. Sensitive detail is gated: the full report goes to buyers who ask. The page changes: updates and subprocessor changes are announced. A small company can follow all four without a single certificate.
What buyers look for on a trust page
A buyer's reviewer reads a trust page with a list in mind. These are the questions they are trying to close, in rough order.
- Who checked this, other than you? A named auditor, tester or tool, independent of the vendor.
- When? A report from two years ago describes another product.
- What was in scope, and what was left out? The production app, or a demo environment.
- What is still open? A vendor that lists open findings is more believable than one that lists none.
- Where is the data, and who else touches it? Hosting region and the list of subprocessors.
- Who do I write to? A security contact, and how you will tell customers about an incident.
- Can I get the full report? Under a confidentiality agreement if needed.
Certifications answer the first three at once, which is why enterprises ask for them. They are also slow and costly for a team of three. The pentest for startups guide covers when a customer contract makes them worth it. Until then, you can answer each question directly.
What a small company can honestly publish without certifications
Write only what is true today, and put a date on it. Six blocks are enough for a first trust page.
| Block | What to write | Example |
|---|---|---|
| What was tested | The live app, by address, and the kind of test | "The production app at app.example.com is attacked from the internet, as a visitor and as two user accounts" |
| When | The date of the last test and how often it runs | "Last full test: 4 November 2026. Tested on every deploy and every week" |
| By what | The tester or tool, independent of you and of your host | "Automated attack test by NAME. No manual penetration test yet" |
| What is open | Counts by severity, and the age of the oldest | "0 critical, 0 high, 2 low open. 7 findings fixed and verified since March" |
| Where data lives | Host, region, subprocessors, each with a link to its own trust center | "Database: Supabase, EU region. Hosting: Vercel. Payments: Stripe" |
| Who to contact | A security address, and a security.txt file | "[email protected]. We tell affected customers about an incident by email" |
The last block has a standard. RFC 9116 defines security.txt, a small file served at /.well-known/security.txt that tells researchers how to report a problem. Two fields are required: Contact and Expires.
Contact: mailto:[email protected]
Expires: 2027-11-13T00:00:00.000Z
Preferred-Languages: enBefore you publish the "what is open" block, make sure you know the answer. Run the same checks a stranger would: tables readable with the public key, secrets in the JavaScript, one user reading another's records. The vibe coding security checklist lists them layer by layer.
What not to claim on a trust page
- "SOC 2 compliant" or "ISO 27001 certified" unless an auditor examined your company and you can hand over the report or the certificate.
- "GDPR certified". Say what you do: where data is stored, your data processing agreement, how to ask for deletion.
- "Bank-level security", "military-grade encryption". These phrases have no test behind them and reviewers discount them.
- "Penetration tested" for an automated scan or test. Name what ran. Buyers who need a manual test will ask who the tester was. The difference is in pentest vs vulnerability scan.
- An undated badge. A seal with no date and no link says nothing about today.
- "No vulnerabilities". Say which tests found nothing, and when. Nobody can honestly claim more.
A false claim on a public page is worse than a gap. Buyers verify, and one claim they cannot confirm puts every other line in doubt.
How to keep a trust page true
A trust page ages the day it is published. Three habits keep it honest.
- 1
Tie the dates to real events
The "last tested" line should change because a test ran. If you update it by hand, you will forget, and a date from last spring tells the buyer the page is abandoned.
- 2
Update the subprocessor list when you add a vendor
Each new tool that receives customer data goes on the list the day you start using it. Large vendors announce those changes, and some contracts require it.
- 3
Show fixes as well as findings
A finding that was open in March and closed in April, with the retest that proves it, is evidence that your process works. Keep the history visible.
What Keybrum's trust page shows, and what it does not
Keybrum is the automated attack test we build for apps made with AI tools. Its plans include a public trust page for each app, which stays hidden until you publish it. It is generated from the tests themselves, so its dates move when a test runs. Here is exactly what a visitor sees on it.
- The app's name and address, and a line stating that the testing is independent, by Keybrum.
- The date of the last full test, the number of attacks it ran, and whether a serious finding is open.
- Open findings counted by severity, the number fixed and verified, the median time to fix, and the date testing started.
- One mark per test over time: green when no attack worked, red when at least one did.
- What the last test tried, by kind of attack and by category of the OWASP Top 10. Which attempts worked is shared only with the app's owner.
- A note on how to read the page, and a link to a printable report with the method, the attacks used, what was not tested and the test history.
The page ends on its own limit: a clean result means these attacks failed on that date, and it is not a promise that no other weakness exists. That sentence is there on purpose. The tests behind the page are attacks on the live app as a stranger: reading tables, storage and database functions with the public key, looking for provider keys in the JavaScript, reading one account's records from another, calling AI routes with no session.
The trust page comes with the plans, from €29 a month excluding VAT (Maker, 5 apps). On the Team plan, €99 a month for 25 apps, the page and the reports carry your own brand, which suits an agency or a company with many internal apps. A one-time report at €49 gives the PDF report without the trust page.
Questions
What is the difference between a trust center and a trust page?
They are the same idea at two sizes. A trust page is one public page with your security facts. A trust center is the larger version: several sections, a document library with gated reports, a subprocessor list and a feed of updates. Start with a page and grow it when buyers ask for more.
Do I need SOC 2 to have a trust center?
No. A trust page is useful as soon as you have true, dated facts to show: what was tested, when, what is open, where data lives and who to contact. SOC 2 becomes necessary when your customers' contracts require it, and then the report goes behind a request on the same page.
What should a startup put on its trust page?
Six blocks: what was tested, when, by what, what is still open, where data is hosted with the list of subprocessors, and a security contact. Add your privacy policy and data processing agreement. Leave out any standard you were not audited against.
Should a trust page show open findings?
Show counts by severity and how old the oldest is, without the technical detail. A buyer trusts a vendor that says "two low findings open, seven fixed" more than one that claims a perfect record. Keep the detail for the report you share on request.
Do I need trust center software?
Not at first. A page on your own site does the job for a small company. Dedicated products earn their price when you have audit reports to gate, many buyers requesting documents and questionnaires to answer at volume.
Does a trust page replace a security questionnaire?
It shortens it. Many reviewers accept a link for the questions the page already answers, and send only the rest. Our guide on security questionnaire answers shows how to answer the testing questions truthfully.
Sources
- 1.OpenAI Trust Portal, OpenAI
- 2.Vercel Trust Center, Vercel
- 3.Security at Supabase, Supabase
- 4.Atlassian Trust Center, Atlassian
- 5.RFC 9116: A File Format to Aid in Security Vulnerability Disclosure, IETF
- 6.OWASP Top 10:2025, OWASP