Security & data handling
Where AnyBot stands on your data (GDPR, a DPA, and residency), plus the agent safeguards, account controls, and an audited trail behind every change. No SOC 2, stated plainly.
AnyBot is built so your data stays yours. Your conversations, knowledge, and prompts are never used to train other models and are never sold or shared. Each space is an isolated environment, access is role-scoped, agents are fenced from your private tools, and every state-changing action is written to a tamper-evident audit log. This page states where we stand honestly, including where a certification isn't in place yet.
Your data
- Never trained on. Your data and prompts are never used to train other models, and never sold or shared with third parties for their own purposes.
- Isolated by space. Agents, contacts, content, billing, and team are scoped to a single space. Nothing bleeds between clients or business units.
- You stay in control. You can export or delete your data, and a DPA is available for teams that need one.
Compliance at a glance
| Area | Where AnyBot stands |
|---|---|
| GDPR | Compliant: data-subject rights (export, deletion) are supported |
| DPA | A Data Processing Agreement is available on request |
| Data residency | Residency options are available; talk to us about requirements |
| Audit trail | Every state-changing action is recorded; see Audit log |
| SOC 2 | Not certified. AnyBot does not hold a SOC 2 report today |
| Encryption at rest | Sensitive credentials are stored encrypted; we don't publish a formal at-rest guarantee yet |
Agent safeguards
The bigger risk with an AI agent isn't the database. It's the model saying or reaching for something it shouldn't. AnyBot fences agents on both sides.
- Grounded answers. Agents answer from the knowledge you give them, not from guesswork, and they're steered away from inventing facts about your business.
- Internals never leak. A redaction backstop strips system prompts, credentials, and internal tool details out of anything an agent says, so a curious visitor can't extract how the agent is wired.
- Public-safe vs owner-only tools. Of the tools an agent can hold, a small public-safe set (27) is all a website visitor can ever trigger; the sensitive owner-only set (95), covering anything that touches billing, team, integrations, or your CRM, is reachable only by authenticated owners and admins, never by an end user in a chat.
- PII redaction on output, and IP addresses are never exposed in the audit trail, a deliberate privacy stance.
Account & access security
- Roles and a permission matrix. Four roles set a baseline, and a per-member matrix of 17 permissions grants or revokes exactly what each teammate can do. See Roles & permissions.
- Owner lock-in protection. A space always has exactly one owner whose access can't be dialed down, so no one can lock the space (or themselves) out.
- Invites are single-use and time-limited, carrying a role and permission snapshot that applies only when the invitee accepts.
Every change is audited
Nothing meaningful happens silently. AnyBot records 559 audited actions (role edits, permission grants, integration changes, agent edits, and more), each with a field-level before/after diff and chained so the log is tamper-evident. That trail is the source of truth for "who changed what, when." See Audit log.
FAQ
Do you train on my conversations? No. Your data and prompts are never used to train other models.
Are you SOC 2 certified? No. AnyBot does not hold a SOC 2 report today. We offer GDPR compliance, a DPA, and residency options instead.
Can a chat visitor reach my CRM or billing? No. Those are owner-only tools, gated to authenticated owners and admins. A public chat can only reach the public-safe set.
Where do I see who changed something? The audit log, which records every mutation with a before/after diff.