How I work: ethical, privacy-first AI
I treat your data, your customers, and your brand like my own.
Data handling
• I only use data you explicitly approve for this project and purpose.
• I minimize sensitive details and do not paste them into public chat tools.
• When possible, I configure zero-retention or private deployments so prompts and outputs aren’t stored or used for model training.
• I do not scrape or ingest third-party content you don’t own or license.
• For demos and testing I use de-identified or synthetic data whenever feasible.
Tool selection and model provenance
• I vet tools for privacy controls, encryption, enterprise terms, and retention settings.
• If your company bans a tool, I propose alternative approaches.
• When feasible, I build inside your accounts or with your API keys so you remain in control.
• Preference for vendors with independent security attestations (for example, SOC 2/ISO) and with clear documentation about training-data sources and training opt-outs.
• For IP-sensitive use cases, I avoid generative tools whose training sources are unclear or likely to include unlicensed material.
Access and security
• Least-privilege access only; I request the minimum permissions required on all project-related accounts.
• Devices are encrypted; secrets and passwords are kept in a password manager—never in plain text.
• Files live in approved folders with encryption at rest and in transit; when sharing.
• If email is unavoidable, attachments are encrypted and keys are shared via a separate channel.
Data lifecycle and deletion
• We agree up front on what I receive, where it lives, who can access it, and for how long.
• I maintain a simple data inventory for the project.
• On project close, I delete local copies and temporary artifacts within the agreed timeframe and confirm in writing.
• Backups and logs follow the same timelines unless your policy requires different handling.
Transparency and approvals
• Every workflow ships with a plain-English one-pager: inputs, steps, tools used, privacy notes.
• Any external integrations, automations, or data connections require written approval.
• I align with your legal and IT policies and adjust after your review.
• A current list of tools and sub-processors used on your project is available on request.
Human oversight and safety
• AI outputs are drafts until a human approves them; nothing auto-sends to customers without your sign-off.
• I test for prompt risks, hallucinations, and data leakage before recommending production use.
• Where relevant, I add guardrails.
• I monitor outputs in pilots and adjust prompts, data, or workflows when necessary.
Copyright and IP
• I avoid ingesting third-party content you don’t own or license.
• Deliverables, custom GPTs, prompts, and workflow documents created for your business are yours as defined in our agreement.
• I do not reuse your proprietary datasets, or workflows in other client work.
• I won’t cross-pollinate competitive strategies between clients without explicit permission.
Marketing use and confidentiality
• I do not use your name, logo, or results in marketing without written permission.
• Testimonials and case studies are drafted for your approval before publication and can be anonymized.
Incident response
• If I detect a data exposure or security issue, I notify your point of contact promptly with facts, scope, and next steps.
• We pause affected workflows until risks are addressed, then document the fix.
• I keep a simple incident log for transparency.
Compliance alignment
• I am not a law firm; I flag issues early and follow your counsel’s guidance.
• If needed, we can add a short data-processing addendum naming approved tools/sub-processors and regions.
• For regulated data (for example, PHI, PCI, or government identifiers), we either put additional controls and agreements in place or avoid using it altogether.
• If your policies require specific data residency, we configure regional controls where vendors support them.
Client choices and controls
• You can require zero-retention modes where supported and available.
• You choose data-residency preferences when tools support region selection.
• You can opt out of any nonessential analytics or logging for assistants I build.
• You approve the data sources a workflow may read and the channels it may write to.
Contact
Questions or requests about privacy and security: douglas@legacymaven.ai
Last updated: August 12, 2025 (HST)