Responsible AI Policy
Effective date: 1 August 2026 Last updated: 1 August 2026
This policy explains the principles applied when Ravi Kumar designs, configures, demonstrates or uses AI-enabled systems in professional work.
1. Purpose before technology
An AI feature should support a defined business purpose. AI should not be added only for appearance, novelty or marketing language.
2. Human responsibility
AI systems do not assume legal or professional responsibility. People remain responsible for:
- Defining the objective
- Approving source material
- Reviewing important output
- Handling exceptions
- Monitoring use
- Making high-impact decisions
- Meeting legal and industry duties
3. Accuracy and limitations
AI output may be inaccurate, incomplete, biased, outdated, inconsistent or fabricated.
A suitable implementation may require:
- Approved knowledge sources
- Citations or traceability
- Human review
- Confidence or uncertainty handling
- Testing
- Restricted actions
- Escalation
- Logging
- Ongoing monitoring
No system should be represented as perfectly accurate.
4. Privacy and data minimisation
Only information reasonably necessary for the task should be processed.
Confidential, personal or proprietary information should not be entered into a third-party AI service without appropriate authority, a valid purpose and suitable safeguards.
Where practical:
- Remove unnecessary identifiers
- Limit access
- Use approved accounts
- Review retention and training settings
- Avoid public consumer accounts for sensitive work
- Document external providers
- Delete temporary data when no longer needed
5. Transparency
Users should be told when they are interacting with an AI agent where that fact is material to trust, safety or decision-making.
A system should not impersonate a real person or falsely claim human review.
6. Human escalation
AI agents should provide a path to a person when:
- The request is outside scope
- Confidence is low
- The user disputes an answer
- Sensitive information is involved
- The matter has legal, financial, medical, safety or reputational impact
- The user asks for a person
- A required integration fails
7. Security and access
AI-enabled workflows should consider:
- Authentication
- Role-based access
- Secrets management
- Prompt-injection risk
- Data leakage
- Tool permissions
- Rate limits
- Abuse prevention
- Audit logs
- Incident response
- Vendor dependencies
An agent should receive only the permissions necessary for its role.
8. Prohibited or unsuitable uses
I may decline work intended to:
- Facilitate fraud
- Deceive or impersonate
- Steal credentials or data
- Create malicious software
- Circumvent lawful controls
- Harass or exploit people
- Produce unlawful discrimination
- Make unsupervised high-impact decisions
- Generate prohibited content
- Violate platform rules or third-party rights
9. Intellectual property
Clients should provide or approve materials they have the right to use. AI-generated material may have uncertain ownership, originality or similarity issues and should be reviewed before commercial publication.
10. Third-party AI providers
AI models and platforms are third-party services with changing:
- Terms
- Pricing
- Availability
- Models
- Output behaviour
- Retention
- regional support
- safety controls
- API limits
No continued availability of a particular provider or model is guaranteed.
11. Testing
Testing should reflect the use case and include:
- Normal requests
- Ambiguous requests
- Incorrect assumptions
- Missing information
- Prompt injection
- Restricted requests
- Escalation
- Integration failure
- Privacy-sensitive inputs
- Unexpected user behaviour
12. Monitoring and improvement
Production systems may require ongoing review of:
- Accuracy
- Escalations
- Failure cases
- Costs
- Response time
- User feedback
- Source changes
- Provider changes
- Security events
13. Client responsibilities
The client remains responsible for:
- Legal and regulatory review
- Accuracy of supplied information
- Industry-specific requirements
- End-user notices
- Staff training
- Access decisions
- Human oversight
- Acceptable use
- Deployment approval
14. Contact
Questions about responsible AI work can be sent to work.ravijethani@gmail.com.