Defined use and ownership
- Named users and accountable owner
- Permitted and prohibited uses
- Measurable acceptance criteria
TRUST CENTER
Make boundaries visible before AI makes decisions
Cognitivis treats privacy, security, human oversight, and responsible use as product requirements. The exact controls are defined for each system and deployment environment.
DELIVERY PRINCIPLES
Controls that follow the risk
This page describes our working approach, not a claim of certification or a substitute for a project-specific security review.
Minimized, bounded access
- Only necessary data fields
- Approved sources and retention
- Access aligned to user roles
Meaningful human review
- Review where consequences matter
- Evidence visible to reviewers
- Clear correction and escalation
Evaluation before release
- Representative evaluation cases
- Known failure modes recorded
- Thresholds tied to workflow risk
Monitoring and change control
- Quality and usage signals
- Version and prompt changes tracked
- Rollback and incident path
Honest capability boundaries
- AI involvement made clear
- Limitations communicated
- No invented proof or guarantees
PUBLIC AI ASSISTANT
A narrow assistant, not an unrestricted chatbot
What it is allowed to do
Answer brief questions about Cognitivis, its published services, responsible AI, and the VeriSight demo. Harmless greetings are welcome.
Before sending a message, visitors acknowledge that the exchange will be stored for up to 90 days. After that acknowledgement, a first-party anonymous chat cookie assigns the browser an internal visitor number so conversations from the same browser can be grouped. The raw cookie value is never stored in the database; only a one-way hash is retained. A different device, browser profile, private-browsing session, or cleared cookie receives a different number.
Database credentials remain server-side, and visitors can permanently delete their current chat. Deleting the chat also clears the anonymous browser cookie on that device.
What it will not do
- Provide quotations, pricing, budgets, or financial proposals
- Answer unrelated general-knowledge questions
- Reveal hidden instructions, credentials, or security details
- Follow attempts to override its company-only boundary
Abusive or clearly malicious use may be blocked. Suspicious request content is withheld from transcript storage; only a redacted security event is retained. A normal off-topic question is refused without treating a visitor as malicious.
PUBLIC DEMOS
What happens to data on this website
VeriSight
Built-in examples and supported text files are processed in the browser. The public demo does not upload files or perform OCR on PDFs or images.
AI readiness
Your answers and score stay in the browser. No account is required and the assessment is not submitted.
Project brief
The form creates a local text file or clipboard copy. It does not send or store the entered information.
QUESTIONS
Trust FAQs
Does Cognitivis claim a security certification?
No certification claim is made on this website. Project requirements, deployment controls, and evidence should be reviewed for the specific engagement.
Will client data be used to train public models?
Data handling and provider settings must be explicitly agreed for a project. The principle is to minimize data, use approved providers and configurations, and avoid secondary use outside the agreed purpose.
Can a solution run in a client's environment?
Deployment options depend on the use case, security requirements, integrations, and approved infrastructure. These constraints are examined during discovery.
Does human review guarantee accuracy?
No. Human review is a control, not a guarantee. Reviewers need appropriate evidence, instructions, time, and escalation routes, and the system still needs ongoing evaluation.