Sanket is an academic and research-oriented AI project focused on Indian Sign Language (ISL) recognition. Because the project interfaces with computer vision pipelines and will eventually process camera/video streams, security, privacy, and responsible data governance are core priorities from day one.
Security and vulnerability updates are applied to active development releases:
| Version | Supported | Status |
|---|---|---|
0.1.x |
✅ | Active Development |
< 0.1 |
❌ | Unsupported |
Do NOT report security vulnerabilities, sensitive data leaks, or privacy issues through public GitHub Issues or Pull Requests.
If you discover a security vulnerability or privacy concern in Sanket, please follow this protocol:
- Email the Maintainers: Send a confidential report to
anveshna2025@gmail.comwith the subject line[SECURITY] Vulnerability Report - Sanket. - Include Key Details:
- Description of the vulnerability or privacy defect.
- Clear steps to reproduce the issue (proof-of-concept script, configuration, or environment details).
- Potential impact on users, video capture feeds, or host systems.
- Any suggested mitigations or patches.
- Encrypted / Secure Channel: If you require encrypted communication, mention this in your initial message to coordinate PGP keys.
To protect researchers, contributors, and sign-language community participants, never publicly post or commit:
- Camera recordings, participant video footage, or raw sign language recordings containing identifiable biometric or facial likenesses without verified consent and formal release documentation.
- Private datasets or copyrighted annotations with restricted redistribution clauses.
- API keys, access tokens, cloud credentials, or private configuration files (
.env, service account JSONs). - Vulnerability exploit details or zero-day demonstration payloads prior to maintainer review and patching.
When a security report is submitted:
- Acknowledgment: The maintainers will acknowledge receipt of the report within 48 hours.
- Assessment & Verification: The vulnerability will be evaluated, reproduced, and prioritized within 5 business days.
- Remediation & Patching: A patch or workaround will be developed on a private security branch and validated.
- Coordinated Disclosure: Once a resolution is merged, a security advisory will be published, and appropriate credit will be attributed to the reporter (unless anonymity is requested).
- No Remote Streaming Without Consent: Sanket is architected to perform client-side or on-device inference; raw video frames are not transmitted over external networks without explicit user authorization.
- No Unencrypted Storage of Video Feeds: Raw camera frames used for landmark extraction should be processed in memory and discarded unless the user has opted into an explicit dataset collection session with documented consent.
- Synthetic & Normalized Features: Where possible, downstream recognition models should consume abstract landmark coordinates (normalized geometry) rather than raw facial or bodily imagery to minimize biometric exposure.