Security
How Premortem is built, what protects your data, and how to report a problem. This page covers the app’s security practices. For a plain-language account of what leaves Jira and what does not, read Where your data goes; the governing legal text is the privacy policy.
Last reviewed: 27 October 2026.
Reporting a vulnerability
security contact: support@usepremortem.com
Send anything you believe is a security issue there — a monitored mailbox on our own domain, not a personal address. Please include what you did, what you saw, and the Atlassian site and app version if you have them.
| Acknowledgement | within 2 business days |
| Initial assessment | within 5 business days |
| Fix for a confirmed high-severity issue | deployed as a Forge release as soon as it is validated, and reported back to you |
We will not pursue anyone who reports a finding in good faith, and we will credit you if you want to be credited. Please do not test against other customers’ sites, and do not run load or denial-of-service tests.
Vulnerabilities in the Atlassian platform itself belong to Atlassian’s own bug bounty programme, not to us.
Architecture: there is no server of ours
Premortem is a Forge app. All of its code runs inside Atlassian’s own infrastructure, under the installation on your site. We operate no servers, no database, no logging pipeline and no network path that receives your content.
That is the single most important security property of this app, and it is structural rather than promised: there is no system of ours to breach, because there is none. It also means we cannot read your issues, your code or your settings — not on request, not under support access, not at all.
The consequences:
- No shared storage. Every Atlassian site gets its own Forge installation with its own isolated storage. One customer’s data is not addressable from another’s.
- No employee access to customer data. There is no console, no admin backdoor and no production database to grant access to, so there is no access to review, revoke or audit.
- Platform security is Atlassian’s. Hosting, tenancy isolation, patching of the runtime and physical security are Atlassian’s responsibility under the Atlassian Trust Center, and Forge apps inherit it.
Data handling
What the app holds
| Data | Where it lives | Deleted |
|---|---|---|
| Settings | Forge storage, your installation only | on uninstall |
| Your AI provider API key | Forge encrypted secret storage | when you replace it, or on uninstall |
| Imported incidents and postmortems, and their embedding vectors | Forge storage, your installation only | on uninstall |
| Risks recorded by your team, including the Atlassian account ID of whoever recorded one | Forge storage, your installation only | on uninstall |
| Who accepted or dismissed a risk, including their Atlassian account ID | a property on that Jira issue, inside your site | with the issue |
Issue text, page text and code diffs are processed and never stored by the app. They are read, sent to the AI provider you configured, and dropped.
Encryption
- In transit: every outbound request is HTTPS/TLS. There is no plaintext transport anywhere in the app, and no option to disable it.
- At rest: Forge storage is encrypted at rest by Atlassian.
- Secrets: your provider API key is written to Forge’s encrypted secret
storage, which is write-only to the app — it is sent in the
Authorizationheader of the request to your provider and is never returned to the user interface, never written to an issue, and never logged.
Where data is allowed to go
The app’s manifest declares an egress allowlist, enforced by the Forge platform rather than by our code. Outbound traffic is only possible to these four hosts:
api.openai.comapi.anthropic.comgenerativelanguage.googleapis.comapi.voyageai.com— optional, only if an administrator chooses Voyage for embeddings
The app cannot reach anything else, including any host of ours. Which of those hosts is used, and whether any is used at all, is the administrator’s choice, authenticated with the customer’s own key. That provider is a data processor under your agreement with them.
You can verify the allowlist yourself: it is the permissions.external.fetch
section of the app’s manifest, which Atlassian validates on every deployment.
Least-privilege scopes
The app requests only the scopes it uses, and Atlassian shows them to an administrator before installation:
read:jira-work, write:jira-work, search:confluence,
read:space:confluence, read:page:confluence,
read:content-details:confluence, write:page:confluence, storage:app.
There is no user-impersonation scope, no administration scope and no scope for reading Jira or Confluence configuration beyond the above.
Access control inside the app
- Everything that changes configuration or spends money — the settings page, connection tests, incident and postmortem imports — verifies that the caller is a Jira administrator on the server side, in the resolver, not only by hiding a button.
- Restricted Confluence pages are never imported. The app reads Confluence with its own rights, so before importing a page it checks that the page and every page above it are open to view; a restricted page is skipped. This exists specifically so the app cannot be used to surface content past a restriction.
- The optional CI integration uses a Forge web trigger. Its URL is itself a credential, and the app can regenerate it, and rotate the signing secret, from the settings page at any time without redeploying.
Secure development
- Automated tests run against the full resolver surface before any release, including the permission and licence checks described above (582 tests at the time of writing).
- Static checks:
forge lintand ESLint must pass; the Forge platform independently rejects a deployment whose manifest violates its rules. - Dependency hygiene: the app ships a small, pinned dependency set, which is reviewed and updated on each release. The Forge runtime itself is patched by Atlassian.
- Release path: code reaches customers only through
forge deployto the production environment, which Atlassian signs and distributes. There is no other way to ship code into the app, and no mechanism for us to push code to one customer and not another. - No telemetry. The app sends no analytics, crash reports or usage data anywhere.
Incident response
If we become aware of a security incident affecting the app, we will investigate immediately, publish a fix as a Forge release, and notify affected Marketplace customers by email at the technical contact on their installation, with what happened, what data was involved and what to do — within 72 hours of confirming it, and sooner if the information is available sooner.
Because the app holds no copy of your data outside your own Atlassian site, there is no dataset of ours that an incident could expose.
Sub-processors
| Who | What for |
|---|---|
| Atlassian | hosting and execution of the app, and all storage of app data, inside your site’s own Forge installation |
| The AI provider you configure — OpenAI, Anthropic, Google, or Voyage for embeddings | processing the text you send for assessment, under your own agreement and your own key |
There are no others. We do not use a CDN, an analytics provider, an error tracker or a hosting provider of our own, because there is nothing of ours to host.
Certifications
Premortem holds no independent security certification. We are not SOC 2 or ISO 27001 audited, and we will not claim to be. We would rather tell you that than imply otherwise.
What we can point to instead is that the app stores no customer data outside the customer’s own Atlassian site, and that the platform it runs on — Atlassian Cloud — is itself SOC 2 Type II and ISO/IEC 27001 certified, which is where your data actually lives. Atlassian publishes its reports at the Atlassian Trust Center.
Data deletion
Uninstalling the app removes its Forge storage, and with it the settings, the API key, the imported incidents and postmortems and their vectors, and the recorded risks.
Two things survive on purpose, because they are yours and are inside your site: risks already written into Jira issue descriptions or into your Confluence risk register, and the per-issue property recording who accepted or dismissed a risk. Delete the issue or the page and they go with it. If you want them cleared while keeping the issues, write to support@usepremortem.com and we will tell you exactly which property to remove — we cannot do it for you, because we have no access to your site.
Questions about anything on this page: support@usepremortem.com.