Prompt-injection defense
Untrusted content is rendered as data, not commands. High-risk tools sit behind human approval, so an injection still cannot move money or push code.
Built so your credentials never touch the AI, every sensitive action waits for approval, and your data never trains a model.
Get Started for FreeViktor is published in the official Slack App Directory. That means our OAuth scopes, security posture, and store listing have been reviewed and approved by Slack before we were allowed to ship to customers through their store.
One-click install from inside Slack. No infra to provision. No long procurement detour just to start a pilot.
Compliance
The audit reports are real, the controls are continuously monitored, and the next audit is always on the calendar.
SOC 2 Type 1
Status
CertifiedCoverage
Independent attestation that our security controls operate as designed. Type II in progress.
Documentation
Report available under NDA.
GDPR
Status
AlignedCoverage
EU data protection requirements met.
Documentation
DPA available on request.
CCPA
Status
CompliantCoverage
California Consumer Privacy Act requirements met.
Documentation
Privacy documentation available.
CASA Tier 3
Status
CertifiedCoverage
Cloud Application Security Assessment, the highest tier required for Google API access.
Documentation
Attestation included in compliance pack.
Slack App Directory
Status
ListedCoverage
OAuth scopes and security posture vetted before shipment through the Slack store.
Documentation
Public App Directory listing.
ISO 27001
Status
In progressCoverage
ISMS controls implementation and evidence collection in progress.
Documentation
Controls overview available today; audit evidence shared after certification.
Data handling
The audit reports are real, the controls are continuously monitored, and the next audit is always on the calendar.
AI Safety
A backend tool gateway injects your API keys and OAuth tokens at execution time. The AI model itself never sees them.
Not a policy. The architecture.
Viktor vs AI tools
AI coworkers introduce attack surfaces traditional SaaS does not have. Three controls keep the surface small.
Untrusted content is rendered as data, not commands. High-risk tools sit behind human approval, so an injection still cannot move money or push code.
Inference runs on OpenAI, Anthropic, and Google. Each is on the public sub-processor list with a no-training agreement for Viktor traffic.
Memory is scoped to your workspace, encrypted at rest, never used to train models, and fully exportable or deletable on request.
Credentials & secrets
Every major tool connects via OAuth with the narrowest scopes that get the job done. No passwords stored.
Where API keys are required, they are stored in a secrets vault, AES-256 at rest, isolated from model context, access-logged, and rotatable.
Admins decide which integrations are connected, who can use them, and at what level. Revoke any integration in one click.

Responsible disclosure
We would rather hear about an issue from a researcher than read about it on Twitter.
We are building a formal bug bounty program. In the meantime, we recognize meaningful security research with a thank you, public credit if you want it, and Viktor credits.
No. Credentials are stored in encrypted vaults and injected at execution time by a backend tool gateway. The AI model never sees them in any context, including planning, execution, or logs.