Security
How Olaa keeps your accounts safe.
Four rules we built the product around: secrets stay in a vault the model cannot read, irreversible actions wait for a human, every workspace is isolated at the database, and nothing you store trains a model.
Where each program stands
| Standard | Status | Coverage |
|---|---|---|
| SOC 2 Type 1 | Audit scheduled | Independent attestation of security controls. Type 2 follows. |
| GDPR | Aligned | DPA and sub-processor list available on request. |
| CCPA | Compliant | Privacy documentation available. |
| Slack App Directory | In review | OAuth scopes and security posture reviewed by Slack. |
| ISO 27001 | Planned | After SOC 2 Type 2. |
What Olaa will and won’t do with your data
- ✓ Encrypts everything: TLS 1.2+ in transit, AES-256 at rest.
- ✓ Stores credentials in an encrypted vault; the model never sees them.
- ✓ Isolates every workspace with database-enforced tenant boundaries and its own sandbox.
- ✓ Logs every tool execution, approval and admin change.
- ✗ Train on your data.
- ✗ Read channels you did not invite Olaa to.
- ✗ Act on anything irreversible without approval.
- ✗ Share anything across workspaces.
Prompt injection
Web pages, emails, documents and tool results are wrapped and labeled as untrusted data before the model reads them, and the system prompt tells Olaa never to take instructions from that content. Even if something slips through, irreversible tools still wait for a human. An injection eval runs against every release.
Found a hole? We want to hear about it.
Write to security@olaa.com. Valid reports earn a thank-you, public credit if you want it, and Olaa credits. The disclosure page has scope and safe-harbor terms.