group: document-parsing-json-structured-data
Data Leakage
Data leakage refers to situations where sensitive information is inadvertently exposed or improperly accessed, leading to potential breaches of confidentiality and security. This can occur through various means such as inadequate data handling practices, misconfigured systems, or unauthorized access.
Key Points
- Sensitive data must be protected at all levels, including during storage, transmission, and processing.
- Data leakage can lead to significant legal and reputational damage for organizations and individuals.
- Best practices include encryption, access controls, and regular audits to prevent leaks.
- “Shadow AI” (unsanctioned AI projects within corporate environments) causes data leakage through lack of oversight, improper data handling, and undocumented data flows.
Local AI Privacy Risks and Mitigation Strategies
Running AI on Your Machine Does Not Make It Private
- The video by [[entities/daniel-miessler|Daniel] Miessler highlights that local execution does not guarantee data isolation if models are trained on leaked datasets or if telemetry is transmitted.
IBM AI Risk Framework & Career Risks
Per Summary Report: IBM’s Five AI Risks & Career-Ending Data Leakage, Martin Keen (IBM Technology) outlines five critical AI risks that can result in termination:
- Data Leakage via Prompting: Accidental inclusion of PII, proprietary code, or confidential business data in public or unsecured AI prompts.
- Unauthorized Data Sharing: Using unsanctioned tools to process organizational data, violating data residency or compliance policies.
- Intellectual Property Exposure: Feeding proprietary algorithms or trade secrets into models that may be used for training or have unclear IP retention policies.
- Hallucination-Driven Errors: Relying on unverified AI outputs for critical decisions, leading to operational failures or compliance breaches.
- Lack of Transparency: Failing to disclose AI use in workflows where human judgment is mandated, undermining accountability and audit trails.