Security & Trust
Acts fast. Stays trustworthy.
Corunner is enterprise software, not a hosted assistant. Human approval is the default on high-stakes actions. Your memory belongs to you. Your data never trains shared models. This page describes how concretely.
A note on certifications
Built to the standards. Pre-certification, but not pre-architecture.
Corunner is engineered to the SOC 2 Type II and GDPR frameworks from day one — the controls, the architecture, and the processes described on this page are in production today. Formal third-party audits are scheduled as part of our public-launch roadmap. If you want the full architecture review now, including DPA template and questionnaire responses, email security@corunner.ai.
How Corunner stays safe
Trust by construction.
Human-in-the-loop, by default
Every action Corunner takes that touches a production system, a customer, or a person routes through an approval card with rationale and citations. Acts fast. Stays accountable.
- Per-action approval policy, configured at onboarding
- Rationale + sources surfaced on every proposed action
- One-click Approve / Edit / Reject in Slack, Teams, Chat, or email
- Auto-deny on stale context after a configurable timeout
Your memory, your control
Organizational memory is yours. Encrypted at rest with per-tenant keys. Permission-aware reads honor your existing access controls. Exportable on request.
- Per-tenant encryption keys (AES-256 at rest)
- Access controls inherited from source systems
- Full export in machine-readable format
- Per-employee redaction and right-to-be-forgotten
No training on your data
Multi-model routing across frontier and open-source models. Customer data is never used to train shared models guaranteed contractually with every model provider on the routing path.
- Contractual no-training guarantees with every routed model
- US-hosted inference for open-source paths
- Model selection auditable per request
- Routing constraints configurable for regulated workloads
Architecture
Designed so the safe path is the default path.
The Corunner architecture is opinionated about safety. Every request inherits permissions. Every action is auditable. Every model route is constrained.
Permission-aware reasoning
When Corunner reasons over your stack, it queries source systems with the actor's own credentials and access boundaries. Corunner cannot read what the operator cannot read.
Multi-model routing
Tasks are routed to the right model — frontier or open-source, US-hosted — based on sensitivity, latency, and cost. Routing decisions are auditable and configurable.
Encryption everywhere
TLS 1.3 in transit. AES-256 at rest with per-tenant keys. Customer-managed keys (CMK) supported on Enterprise.
Tenant isolation
Strict logical isolation between tenant memories, embeddings, and inference contexts. No cross-tenant prompts, no shared caches.
Continuous observability
Every Corunner action emits structured audit events to your SIEM. Real-time anomaly detection flags unusual access patterns to the admin web app.
Safe defaults
Read-only mode by default on new integrations. Write access is opt-in, scoped per connector, and surfaces in every approval card.
Compliance posture & roadmap
Where we stand, line by line.
Capabilities marked Available now are in production today. Certification frameworks marked Built to are architectural commitments engineered into the product; formal third-party audits are scheduled and progress can be shared on request.
Data handling
What we keep. Where we keep it. How you delete it.
What we store
Organizational memory entries (decisions, owners, terminology, history) and minimal operational metadata. Source-system content is referenced, not duplicated, wherever possible.
Where we store it
All storage and inference run in the US region today. EU residency is on our near-term roadmap; we will notify customers ahead of opening it. Storage and inference are always co-located in the chosen region. Customer-managed keys available on Enterprise.
How long we keep it
Memory persists for the lifetime of the contract by default. Per-entry and per-employee retention policies configurable through the admin web app.
How to delete it
Single-employee redaction, per-entry deletion, and full-tenant export-and-purge are all available from the admin web app. Deletes propagate end-to-end within 30 days.
Responsible disclosure
Report a vulnerability.
Found something? We respond to coordinated disclosure within one business day. We don't pursue legal action against good-faith research.
PGP fingerprint: Available on request.
Acts fast. Stays trustworthy.