A customer needs answers
Use the MSA, AI addendum or security questionnaire to identify the commitments behind the review.
For growing AI companies · Design partners welcome
Connect your customer commitments and relevant security controls to checks across your cloud, code and AI workflows. Find gaps, work through fixes, and verify what changed.
Start with customer documents or recommended checks. Documents are optional.
From a customer promise to work delivered
Illustrative workflowThe commitment
“Customer prompts must not persist in application logs.”
The work
Review potential payload logging. Prepare a targeted code change and regression test.
The evidence
After deployment, test the reviewed paths and record the result, scope and remaining follow-up.
The work on your plate
Your agreements and policies describe how you protect data and control AI. As the product changes, those commitments need evidence and follow-through.
Use the MSA, AI addendum or security questionnaire to identify the commitments behind the review.
Start with a concern your engineers already have. Connect the relevant systems and work toward a verified result.
Choose recommended checks without uploading a document. See the scope, findings and work that still needs attention.
How it works
Agree on the work, connect the sources it needs, and keep each finding moving toward a clear outcome.
Start with customer commitments or recommended checks. Review the work and approve the access it needs.
See observed gaps, potential issues and missing evidence, with a source and a clear next action.
Review a supported code change as a pull request, or follow targeted configuration guidance.
After your team applies the change, collect fresh evidence of what improved and what remains unresolved.
Save a dated report connecting the source, checked systems, changes and remaining work.
Repeat approved checks on a schedule and keep follow-up assigned to an owner.
Starting coverage
Start with a supported concern we can investigate and act on. Each result states the systems, resources and evidence it covers.
Check the setting behind the promise
“Are our customer exports protected from accidental overwrite?”
Read the selected storage setting, identify a version-history gap, and verify the configuration after your team fixes it.
Scope today
AWS S3 versioning on one selected bucket. Backup coverage and restore testing need separate evidence.
Find a gap. Work through the fix.
“Are customer prompts or model outputs ending up in our logs?”
Scan selected application logs for potential sensitive data. Review an applicable logging fix, approve its pull request, then verify the deployed mitigation with fresh tests.
Scope today
CloudWatch Logs and a supported Python logging pattern. Model-provider retention, training and other destinations need separate evidence.
Control a consequential action
“Can our agent make this change before a person approves?”
Set trusted bounds, hold consequential changes for a named approver, and record the policy decision and actual execution outcome.
Scope today
One configured numeric MCP action through a customer-run gateway. Integration is required; protection applies to the routed action.
Checks draw on relevant security guidance. Unavailable assessments and unsupported commitments stay visible without receiving a passing result.
A record your team can use
Review and save a private, dated report connecting commitments, checked systems, fixes and remaining gaps. Download HTML to read or share, and JSON to retain the structured record.
Reports stay fixed when newer checks run. Signed action records capture the decisions and gateway-reported outcomes for your protected workflow.
What belongs in the record
The source commitment or recommended check, with its scope.
The resource, check, date and evidence behind the result.
Reviewed fixes, customer deployment and fresh verification.
Unknowns, unsupported obligations and owned follow-up.
Evidence supports a scoped review. It is not a certification or a guarantee against breaches.
Ongoing checks
Opt into hourly or daily checks of approved resources, with on-demand rechecks when needed.
Evaluate each routed action before execution. Larger changes wait for the configured human approval.
In-app alerts identify new issues and regressions. Assign an owner and keep the finding separate from its acknowledgement.
Work with us as a design partner
Bring a customer requirement, a known gap or a system you want assessed. We’ll agree on the supported scope, access, responsibilities and success criteria, then work through the gap to a checked result.
Work directly with the founder and help shape the security work Estoppl takes off your team’s plate.
Tina Ho · founder · LinkedIn ↗
No. Choose recommended checks or start from a known concern. You can add customer documents and review new proposed work later without restarting onboarding.
Assessment connections use scoped read access. You approve supported pull requests, and your team merges and deploys them. Agent enforcement is a separate integration with explicit action limits and approval settings.
The result applies to its recorded resources, configuration or tested workflow. Missing, stale and unsupported evidence remains visible. A logging test does not establish model-provider zero retention or complete company-wide security.
Developer resources
The Apache 2.0 proxy provides its own policy controls and signed event trail. These resources cover the proxy and its verification examples; the assessment and repair service has the scope described above.