Skip to main content

The RSG Secure Systems Standard

AI automation without security is a liability.

RSG does not simply connect tools and hope they work. Permissions, approvals, backups, logs, testing, and recovery procedures are designed into every system from the beginning — because these systems run real business operations and hold sensitive data.

Security-first development

A standard applied to every website, CRM, automation, portal, and custom system RSG delivers

RSG evaluates every project across the areas below. Which controls a system receives depends on what it does and the data it holds — security is selected to fit the project, not sold as a fixed bundle.

Security is part of the build

RSG systems are designed with controlled access, secure credentials, backups, audit trails, responsible-AI safeguards, and documented recovery from the beginning. Security is not added after the automation is already running.

AI should not have unlimited authority

RSG can require human approval before an AI system sends sensitive communications, changes records, approves financial actions, releases documents, or performs other high-risk tasks.

Know where your data goes

RSG documents the third-party platforms involved in each system, what they are used for, and what categories of information they may process.

Built for accountability

Important administrative, customer-data, and AI actions can be logged so a business understands what happened, who authorized it, and which records were affected.

The standard, area by area

Ten areas RSG evaluates on every project

Identity & access

The right people, the right access — and nothing more

Employees, administrators, customers, contractors, and vendors should only reach the information and functions their role requires. Access is designed before the system is built, not bolted on afterward.

  • Multifactor authentication
  • Secure account creation and login
  • Role-based access permissions
  • Least-privilege access
  • Session management
  • Account recovery controls
  • Administrative access restrictions
  • Optional single sign-on for eligible projects

Permissions are enforced on the server and, for databases, at the data layer — hiding a button is never treated as the control.

Data protection

Sensitive data is protected and never over-collected

RSG protects data in transit and, where the selected infrastructure supports it, at rest — and avoids collecting or keeping information a system does not actually need.

  • Encryption in transit
  • Encryption at rest when supported by the infrastructure
  • Secure database access
  • Sensitive-field protection
  • Data-retention controls
  • Data-deletion workflows
  • Export and portability controls
  • Restricted access to confidential information
  • Prevention of unnecessary data collection

Less data retained means less to lose. Retention rules are documented, not assumed.

Secrets management

Credentials stay out of the browser and out of source control

API keys and credentials belong in managed secret storage — never in frontend code, never committed to the repository, and never shared across environments.

  • No exposed API keys in frontend code
  • No credentials committed to source control
  • Environment-variable management
  • Managed secret storage where available
  • Restricted production credentials
  • Credential rotation procedures
  • Separate credentials for development, staging, and production
  • Immediate invalidation of compromised credentials

Infrastructure & environments

Development, staging, and production stay separate

Where appropriate, work happens in separate environments. Development data should not automatically contain real customer information, and production must not rely on test credentials, placeholder integrations, mock authentication, or developer bypasses.

  • Separate development, staging, and production environments
  • Real customer data kept out of development by default
  • No test credentials or mock authentication in production
  • Production debugging disabled
  • Demo systems isolated from production data
  • Secure deployment process

Backups & recovery

Recovery is planned, documented, and tested

It is not enough to say backups exist. RSG documents what is backed up, how often, how long it is retained, and how a restoration would actually be performed.

  • Automated database backups
  • Configurable backup frequency
  • Retention schedules
  • Restoration procedures
  • Recovery testing
  • File and document backup considerations
  • Business-continuity planning
  • Defined recovery expectations when in scope

Logging & accountability

A clear record of who did what, when, and to which record

Important administrative, customer-data, and AI actions are logged so a business can understand what happened and who authorized it — without ever writing passwords, keys, or full payment details into the logs.

  • Activity logs
  • Administrative audit logs
  • Authentication events
  • Permission changes
  • Record creation, modification, and deletion events
  • AI-generated action records
  • Integration-failure records
  • Exportable logs when appropriate

Responsible AI controls

AI should not have unlimited authority

For systems that generate content, make recommendations, access business data, or communicate with customers, RSG can require a human to approve high-risk actions before anything happens — and keeps untrusted input separate from instructions.

  • Human approval for high-risk actions
  • Permission-aware document access
  • Restricted tool access
  • Approved data-source boundaries
  • Prompt-injection and disclosure testing
  • Output validation and structured outputs
  • Rate limits and action limits
  • Escalation to a human
  • Audit trails for AI actions
  • Emergency disable controls

High-risk actions — sending contracts, issuing refunds, changing pricing, deleting records, sharing confidential documents — do not execute automatically without deliberate authorization.

Vendor management

Know where your data goes

RSG documents the third-party platforms involved in each system — hosting, databases, email, SMS, payments, analytics, AI models, storage, and authentication — what they are used for, and what categories of information they may process.

  • Vendor name and service
  • Business purpose and data handled
  • Hosting region when known
  • Security documentation and agreement status
  • Subprocessors
  • Access level
  • Renewal and review dates
  • Replacement or removal procedure

RSG does not automatically claim that any vendor is approved for a regulated industry.

Incident response

A plan for when something goes wrong

Every qualified project includes an incident-response procedure sized to its risk — how an incident is detected, who is responsible, how access is contained, how credentials are rotated, how systems are isolated, how services are restored, and how affected clients are notified.

  • Detection and ownership
  • Access containment
  • Credential rotation
  • System isolation
  • Log preservation
  • Service restoration
  • Client notification
  • Documented lessons and corrective actions

Testing & maintenance

Security is verified, then maintained

AI-enabled projects are tested against an internal checklist — prompt injection, sensitive-information disclosure, permission bypass, data leakage, excessive agency, and more — with results, severity, evidence, and retest status recorded.

  • Prompt-injection and indirect-injection testing
  • Sensitive-information disclosure testing
  • Permission-bypass and unauthorized-retrieval testing
  • Data-leakage testing between customers
  • Excessive-agency and unsafe-tool-call testing
  • Abuse and rate-limit testing
  • Backup restoration testing
  • Periodic security review

Human-in-the-loop

High-risk AI actions wait for a human decision

When an AI feature drafts a high-risk action, it does not execute automatically. It enters an approval queue where a person can approve, reject, edit, or escalate — and every AI action can be logged.

01Drafted by AI
02Awaiting approval
03Approved
04Executed
or Rejected / Escalated

Actions that require deliberate authorization

Sending contracts
Issuing refunds
Changing pricing
Deleting records
Approving estimates
Modifying permissions
Sharing confidential documents
Making financial commitments
Publishing public content
Sending sensitive customer communications
Scheduling expensive work
Executing database or infrastructure changes

Framework-informed development

Informed by NIST and OWASP guidance

The RSG Secure Systems Standard is informed by established security and responsible-AI principles. RSG organizes its internal AI risk process around four stages.

Govern

Decide who owns the AI, what it may do, and how much risk is acceptable.

  • Ownership and responsibilities
  • Acceptable use and risk tolerance
  • Approval requirements
  • Documentation standards
  • Vendor oversight
  • Incident escalation

Map

Understand the intended use, who it affects, and what could go wrong.

  • Intended use and users
  • Data sources and dependencies
  • Affected people and business consequences
  • Potential misuse
  • Regulatory considerations
  • Actions the AI can perform

Measure

Evaluate how well it works and how it fails.

  • Accuracy and reliability
  • Security and privacy
  • Prompt-injection resistance
  • Data-leakage risk
  • Failure behavior
  • Human-approval effectiveness and logging completeness

Manage

Put controls in place and keep them current.

  • Risk controls and restricted permissions
  • Human review and monitoring
  • Incident procedures and remediation
  • Model or vendor changes
  • Feature shutdown procedures
  • Periodic reassessment

OWASP-informed AI risks, in plain terms

Prompt injection

Untrusted input is kept separate from instructions and tested against injection.

Sensitive-information disclosure

The assistant is scoped so it cannot reveal data beyond a user's permissions.

Improper output handling

AI output is treated as untrusted — escaped when displayed, validated before any action.

Excessive agency

Tool access is restricted and high-risk actions require human approval.

Insecure integrations

Third-party connections are documented, scoped, and reviewed.

Unauthorized access

Authorization is enforced server-side and, for databases, at the data layer.

Unsafe dependency or vendor usage

Vendors and dependencies are documented and reviewed.

Unbounded resource consumption

Rate limits and action limits contain cost and abuse.

An honest word on compliance

  • Built using principles informed by NIST and OWASP guidance.
  • Security controls are selected based on project requirements.
  • Designed around secure development and responsible-AI practices.
  • Compliance-ready architecture may be available when specifically scoped.
  • Final compliance obligations depend on the client, industry, vendors, configuration, and legal requirements.

Security packaging

Security scaled to the system

Controls are grouped so the right depth of security fits each kind of project — from a standard website to a confidential private-AI deployment.

RSG Core Security

Standard websites and basic business systems.

  • Secure authentication
  • MFA for administrators
  • Environment-variable secret management
  • Encrypted connections
  • Automated backups
  • Basic activity logs
  • Secure deployment checklist

RSG Advanced Security

CRMs, customer portals, operations systems, and sensitive business workflows.

  • Granular role-based access
  • Detailed audit logging
  • Vendor and subprocessor records
  • Data-retention controls
  • Documented recovery procedures
  • Environment separation
  • Security testing
  • Incident-response documentation

RSG Responsible AI Controls

AI assistants, agents, communication tools, and automated decision workflows.

  • Human approvals for high-risk actions
  • Tool restrictions
  • Permission-aware retrieval
  • Prompt-injection testing
  • Output validation
  • Sensitive-data controls
  • AI action logs
  • Escalation procedures
  • Emergency disable controls

RSG Private AI Security

Organizations with greater confidentiality requirements.

  • Private-cloud or local deployment options
  • Restricted model and vendor access
  • Controlled knowledge ingestion
  • Network restrictions
  • Custom retention policies
  • Advanced access controls
  • Detailed audit logs
  • Confidentiality-focused architecture
  • Client-controlled approval procedures

Pricing is scoped per project. RSG does not publish fixed security pricing, and final compliance obligations depend on the client, industry, vendors, configuration, and legal requirements.

Questions

Frequently asked

Is RSG certified or compliant with SOC 2, HIPAA, or other frameworks?

No. RSG builds using principles informed by NIST and OWASP guidance, but it does not represent itself as certified, audited, or compliant with any framework unless that status has been independently verified for a specific engagement. Compliance-ready architecture can be scoped when a project requires it, and final compliance obligations depend on the client, industry, vendors, configuration, and legal requirements.

Does every system get every control?

No. Security controls are selected based on the project. A marketing website needs secure forms, spam protection, admin authentication, backups, and secure deployment. A CRM adds role-based permissions, audit logs, retention rules, and recovery procedures. An AI operations system adds tool restrictions, human approval, prompt-injection testing, and complete AI action logging.

How does RSG keep AI from doing something it shouldn't?

High-risk AI actions — sending contracts, issuing refunds, changing pricing, deleting records, sharing confidential documents, making financial commitments — do not execute automatically. They enter an approval queue where a human can approve, reject, edit, or escalate before anything happens, and every AI action can be logged.

What happens to my data, and where does it go?

RSG documents the third-party platforms involved in your system, what each is used for, and what categories of information it may process. Data is protected in transit and, where the infrastructure supports it, at rest. RSG also avoids collecting or retaining information a system does not actually need.

Are backups real, or just a checkbox?

RSG documents what is backed up, how often, how long it is retained, and how a restoration would be performed — and restoration is tested rather than assumed. Backup status shown in the admin console reflects real infrastructure or is clearly marked when it cannot be verified from the console.

What if there's a security incident?

Every qualified project includes an incident-response procedure sized to its risk: how an incident is detected, who is responsible, how access is contained, how credentials are rotated, how systems are isolated, how logs are preserved, how services are restored, and how affected clients are notified — with lessons and corrective actions documented.

Next step

Build a system that's secure from the start

Every engagement starts with a strategy call. We review the business and its data first, then design the controls that fit.