Recommended Free Tools
Verify an AI service by documenting the exact product, model, deployment route, configuration, data types, processors, retention rules, user permissions, and independent assurance that apply to your purchase. A security badge or a homepage statement is only a starting point; approval should depend on evidence tied to the service and configuration you will actually use.
What a defensible verification actually proves
Your objective is not to prove that an AI company is secure in the abstract. It is to establish, for a defined service and contract, where your data goes, which tenant boundary protects it, how long each category remains, who can access it, and what records demonstrate those controls.
As an Amazon Associate I earn from qualifying purchases.
Scope every answer. A provider can operate several products, model families, API features, regions, and hosting arrangements with different processors, logs, and deletion behavior. The controls for a direct API call may not be the controls for the same model accessed through a cloud platform. A commercial chat product that saves conversations also has a different data lifecycle from a stateless inference endpoint.
Ask for written evidence and record its version and date. Then compare that evidence with the applicable agreement, data-processing addendum, security documentation, product configuration, and audit report. Provider documentation establishes what the provider states; it does not independently test every implementation or prove that your tenant is configured correctly.
#1 Best Overall
Define the exact service boundary before asking questions
Inventory the product and configuration
Create a one-page scope record before procurement or a security review. Include:
- Product name, edition or plan, and the contracting legal entity.
- Model family and specific model versions, including any automatic model-routing feature.
- API endpoints, assistants, agents, fine-tuning, file search, embeddings, evaluation, or other enabled features.
- Deployment route: direct provider service, private deployment, or access through a cloud platform.
- Region, data-residency selection, cross-border transfers, and connected applications or tools.
- Customer settings for training or service improvement, abuse monitoring, conversation history, exports, and deletion.
- Data categories: prompts, outputs, uploaded files, conversation history, application state, embeddings or indexes, telemetry, support tickets, abuse records, audit logs, and derived metadata.
- Contract, security addendum, and privacy-policy versions that govern the purchase.
Do not accept “the platform” as a sufficient scope description. Put the model, feature, route, region, and plan in every question and in the final approval record.
Map every processor and subprocessor
Ask the provider to identify every organization that handles each data category, not merely the company named on your invoice. Require the list to distinguish primary processing, storage, support, monitoring, analytics, security, and backups.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Routing through another cloud changes the responsibility analysis. Anthropic’s documentation states that, for Amazon Bedrock and Google Cloud’s Agent Platform, the cloud provider is the data processor. Review that provider’s retention, support-access, regional-transfer, and compliance documentation as well as the model maker’s terms. Record which organization is responsible at each step and which contract gives you deletion, incident-notification, and audit rights.
Verify isolation as a control, not a slogan
Request an architecture and data-flow explanation
Ask for a diagram or written narrative that follows a request from ingestion to inference, storage, monitoring, support, backup, and deletion. It should identify production systems, shared services, search indexes, queues, caches, object stores, and disaster-recovery copies. Require the provider to mark where tenant identifiers are enforced and where data is encrypted, decrypted, transformed, or combined with metadata.
Obtain a written explanation of the tenant boundary. Useful answers describe logical or physical separation, authorization checks on every request, service-to-service identity, key-management boundaries, and controls that prevent one customer from querying another customer’s records. Ask how isolation applies to files, conversation history, embeddings, fine-tuning artifacts, support attachments, and backups—not just the model’s immediate input.
Test employee and support access assumptions
Ask how production access is constrained for engineers, operations staff, security personnel, and support agents. The response should cover unique identities, least privilege, approval before privileged access, time-limited elevation, session recording or logging, emergency access, and periodic review. Ask whether support personnel can view prompts, outputs, files, or account metadata, and under what customer-authorized or security-related circumstances.
Request the retention period and customer availability of those access records. A control that is not logged, reviewed, or attributable to an individual is difficult to verify after an incident.
Do not treat encryption as proof of tenant separation
Encryption in transit and at rest protects data while it is moving or stored, but it does not by itself demonstrate correct authorization, tenant boundaries, privileged-access controls, or safe deletion. OpenAI’s business security materials describe encryption, role-based controls, and Enterprise Key Management. Treat those statements as questions to validate against the exact service, plan, key arrangement, and contract you will use—not as proof that every route has identical controls.
Build a retention table for every data type
Request a table rather than a single retention headline. For each category, capture the purpose, default period, customer-configurable period, deletion trigger, backup expiry, legal or safety exceptions, and whether the data may be used for service improvement or abuse monitoring.
| Data category | Questions to answer |
|---|---|
| Prompts and outputs | Are they stored, for what purpose, and for how long? Does deletion remove active records, logs, indexes, and backups? |
| Application state and saved conversations | Which features save state? Can an administrator or user delete it? What remains after deletion? |
| Uploaded files and attachments | Where are files stored, are copies made for retrieval or safety scanning, and when do all copies expire? |
| Embeddings, indexes, and fine-tuning artifacts | Are derived representations retained separately from source data, and does the same deletion request cover them? |
| Abuse-monitoring records | What prompts, outputs, metadata, or flags are collected, who can see them, and what exceptions extend retention? |
| Compliance and audit logs | Which events are captured, how long are they available to the customer, and what data must the provider retain internally? |
| Backups and disaster-recovery copies | What is the backup expiry schedule, and can data be restored after a customer deletion request? |
| Support and incident records | Do tickets or investigations contain customer content, who can access them, and when are they deleted? |
Clarify what “delete” means. It may mean removal from active systems while encrypted backups, security logs, legal holds, or abuse records remain for a defined period. Ask who can invoke deletion, whether the request is synchronous or queued, what confirmation you receive, and how the provider handles data restored from backup.
Examples showing why one duration is insufficient
- OpenAI’s current API data-controls documentation says abuse-monitoring logs may contain prompts, responses, and derived metadata and are retained for up to 30 days by default, unless a stated exception applies. That category is distinct from application state and other stored data. The same documentation says, “As of March 1, 2023, data sent to the OpenAI API is not used to train or improve OpenAI models (unless you explicitly opt in to share data with us).” Treat the statement as applying to the documented API scope and date, not as a promise about every product or route.
- OpenAI’s current Compliance Platform documentation says compliance logs are retained for 30 days. It also states that the API cannot delete some information retained internally for audit or security purposes. Compliance logs therefore need their own row in your retention table.
- Anthropic states that standard commercial API inputs and outputs are automatically deleted within 30 days by default, with exceptions for features that have longer customer-controlled retention, a different agreement such as zero data retention, Usage Policy enforcement, or legal compliance. Commercial products that save conversations retain chats and coding sessions to provide the product experience.
- Anthropic’s covered-model notice specifies a 30-day prompt-and-output retention period for covered models, effective June 9, 2026, and identifies the organizations and zero-retention arrangements affected. A “zero retention” label therefore requires a model-, feature-, organization-, and agreement-specific exception check.
These figures are product- and category-specific: “up to 30 days” for OpenAI API abuse-monitoring logs, “30 days” for OpenAI compliance logs, “within 30 days” for Anthropic’s standard commercial API deletion timeframe, and “30 days” for Anthropic covered-model prompts and outputs. OpenAI’s figures were documented as current on October 13, 2026; Anthropic’s standard API timeframe is dated July 1, 2026, and its covered-model policy takes effect June 9, 2026. Do not convert any of them into a universal promise for an entire provider.
Check provider access and customer access separately
Provider-side access questions
Request a role matrix covering end users, workspace owners, administrators, support staff, engineers, security personnel, and subprocessors. Verify:
- Least-privilege permissions and separation of duties.
- Unique identities, strong authentication, and restrictions on shared accounts.
- Joiner, mover, and leaver processes that remove access promptly.
- Approval, time limits, and monitoring for privileged or emergency access.
- Periodic access reviews and documented remediation of excess privileges.
- Audit events showing who accessed which tenant, when, why, and through what mechanism.
OpenAI’s data-processing terms describe employee access based on least privilege and limit customer-data use to applicable agreements. Confirm that the language covers the service and legal entity in your order, and ask for the operational evidence or audit scope that supports it.
Customer-side access questions
Map your own permissions just as carefully. Identify who can create or rotate API keys, change retention or training settings, invite users, connect applications, export records, delete workspaces, and retrieve compliance logs. Require single sign-on and multi-factor authentication where available, and document emergency recovery if an identity provider is unavailable.
OpenAI’s Compliance Platform documentation says workspace compliance data is accessed with a workspace-scoped Admin key and permissions and identifies audit, authentication, and app logs among supported categories. Verify that your purchased entitlement includes those logs, that the key can be restricted to the required workspace, and that your administrators can retrieve and protect the records without granting unnecessary content access.
Establish log availability and retention on your side. Export events to a controlled system if the provider’s retrieval window is shorter than your regulatory, investigative, or records-management requirement. Test a sample export before production use so you know whether it contains user identifiers, object IDs, timestamps, administrator actions, and deletion events.
Read assurance reports, not just certification badges
Ask for the current SOC 2 report, applicable ISO certificate and statement of applicability, penetration-test summary if available, and any bridge letter or remediation statement needed to cover the current operating period. For each artifact, verify:
- Audited legal entity and service organization.
- Products, model services, regions, and infrastructure actually in scope.
- Control criteria, examination or certification dates, and report period.
- Subservice organizations, carve-outs, complementary customer controls, and exclusions.
- Exceptions, qualified opinions, control failures, and remediation status.
- Whether the report covers the direct service or only a cloud-hosted infrastructure layer.
OpenAI reports an independent SOC 2 Type 2 examination for relevant API and ChatGPT business services and lists ISO certifications. That public statement does not establish that your selected model, route, region, or configuration is covered. Obtain the applicable report or certificate under the provider’s disclosure process and match its scope to your purchase.
Free tools Windows power users keep installed
One-click scans. No signup required.
A practical verification workflow
- Write the boundary. Record product, plan, model, endpoints, features, route, region, connected tools, settings, data categories, contract version, and every known processor.
- Request isolation evidence. Obtain the data-flow description, tenant-boundary mechanisms, employee-access policy, support-access process, backup treatment, and relevant audit evidence. Ask specifically about shared services, search indexes, and connected applications.
- Complete the retention table. Capture defaults, configurable controls, deletion triggers, backup expiry, legal and safety exceptions, service-improvement use, and abuse-monitoring treatment for each category.
- Verify access and logs. Review provider and customer role matrices, authentication, privileged-access approvals, audit events, export functions, and the period for which logs remain retrievable.
- Validate independent assurance. Check the legal entity, service scope, period, subservice organizations, exceptions, and remediation in current SOC, ISO, penetration-test, or equivalent evidence.
- Record the decision. Put every requirement, provider answer, supporting artifact, covered route, version, owner, exception, and approval decision in a control matrix. Assign unresolved items to security, privacy, legal, or procurement owners.
Use one comparison framework for providers and deployment routes
| Comparison axis | Evidence to compare | Decision question |
|---|---|---|
| Data boundary | Direct service or cloud-hosted route, processors, subprocessors, regions, and transfers | Can you identify responsibility for every processing step? |
| Isolation | Architecture, tenant-boundary controls, contractual commitments, and audit scope | Is cross-customer access prevented and independently examined? |
| Retention | Per-category defaults, configurable periods, deletion semantics, backups, and exceptions | Can the service meet your deletion and records obligations? |
| Access | Provider staff roles, customer roles, privileged support, authentication, and logs | Can you limit, detect, and investigate access? |
| Assurance | Audited entity, product coverage, dates, exceptions, and report availability | Does independent evidence cover the service you will operate? |
| Operational fit | Available settings, evidence retrieval, administrative effort, and change notices | Can your team configure and continuously demonstrate compliance? |
Compare like with like. A direct API route with a contractual zero-retention arrangement should not be compared with a saved-conversation product as though they had one common lifecycle. Likewise, a model accessed through a cloud platform must be compared with the cloud platform’s processor terms and support controls, not only with the model provider’s public page.
Common verification failures and how to correct them
Relying on a homepage or badge
Failure: Treating “enterprise security,” an ISO logo, or a SOC 2 reference as proof that every product and configuration is covered.
Correction: Obtain the current report, inspect scope and exceptions, and attach it to the exact service, route, legal entity, and contract version in your matrix.
Using one retention number for all data
Failure: Recording “30 days” without distinguishing prompts, saved chats, abuse logs, compliance logs, files, indexes, and backups.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchCorrection: Require one row per data type, including deletion behavior and exceptions. Mark “not stated” where the provider has not established a value rather than filling the gap with an assumption.
Assuming deletion removes every copy immediately
Failure: Approving a service without asking about backups, legal holds, safety investigations, internal audit records, or restored data.
Correction: Define active-system deletion, backup expiry, exceptions, confirmation evidence, and the process for data restored after deletion.
Checking only provider employees
Failure: Reviewing vendor staff access while leaving your own administrators, API keys, integrations, and export permissions unmanaged.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Clear out junk files and repair common Windows errors3Fix the driver behind crashes, sound loss and screen glitchesCorrection: Maintain separate provider and customer role matrices. Enforce SSO and MFA, rotate and scope keys, review privileged roles, and test audit-log retrieval.
Ignoring the hosting route
Failure: Assuming that a model provider’s direct-service terms govern a model called through another cloud.
Correction: Identify the processor for each step and review the cloud platform’s retention, support, regional, and compliance documentation as part of the same approval.
Keep the evidence current after approval
Make the control matrix a living record. Recheck it after a material model release, new feature, hosting-route change, region change, contract amendment, retention-policy revision, or subprocessor change. Require the provider’s change-notification process to reach the owner responsible for privacy and security review.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Schedule periodic checks of report periods, certificates, bridge letters, privileged-access evidence, deletion tests, and log exports. If a provider cannot answer an isolation question, define retention precisely, supply applicable audit material, or identify processor responsibility, record the gap explicitly and escalate it rather than converting an unverified assertion into an approval.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
