Resources · Technical References
Top 10 End-User LLM Security Precautions
Practical guidance for individuals and organizational users of generative AI systems. Follow your organization's approved-use and data-handling policies.
- Use approved AI services. Verify enterprise approval, terms and administrative controls before using work data.
- Classify information before prompting. Never paste secrets, credentials, regulated data or confidential records without explicit authorization.
- Minimize data shared. Remove unnecessary identifiers; anonymization must be evaluated for reidentification risk.
- Understand retention and training settings. Review service terms, logging, retention and model-improvement policies.
- Verify outputs. Independently check factual claims, calculations, citations and generated code before relying on them.
- Treat model output as untrusted. Scan generated code, commands and links; never execute blindly.
- Watch for prompt injection. Documents, webpages, emails and tool results can contain malicious instructions.
- Restrict integrations and agent permissions. Use least privilege, limited scopes and approval gates for consequential actions.
- Protect accounts and sessions. Use MFA, secure devices and organizational identity controls.
- Report incidents promptly. Escalate suspected disclosure, unsafe output or unauthorized agent activity through established channels.
Framework relationship
These precautions support the IAIGSRM readiness and control-delivery approach. Formal control identifiers and verified standards mappings will be introduced only after technical review.