Back Mitigate

Security Policy

Mitigate Procurement AI Agents software

v1.3, February 2026

Introduction

This Security Policy outlines the measures and protocols in place to ensure the security and integrity of the Mitigate Procurements Platform. The platform provides AI-powered analysis of procurement documents (RFPs) against vendor proposals, automatically detecting compliance gaps, mismatches, and risks. This policy is designed to protect the platform, its data, and its users from unauthorized access, disclosure, alteration, and destruction.

Scope

This policy applies to all clients, employees, contractors, and third-party service providers who have access to the Mitigate Procurements Platform or its data. It covers the web application, background processing systems, AI analysis agents, document storage, and all integrations with third-party services.

Platform High-Level Architecture

Mitigate Procurements Platform is a Ruby on Rails 8 cloud application with a reactive user interface powered by Hotwire (Turbo and Stimulus). Customer data is stored in secure PostgreSQL database instances on Amazon Web Services (AWS). All data in the platform is fully encrypted, both in transit and at rest.

Our web infrastructure is hosted on Heroku, a world-class PaaS service, with resources hosted within the EU region. This setup provides strong security, regulatory alignment with GDPR, and the capacity to grow with our customers. More about Heroku's security: https://www.heroku.com/policy/security

The platform employs a two-tier AI agent architecture for document analysis:

  • Main Orchestrator Agent — handles reasoning, orchestration, and tool selection
  • Specialized Sub-Agents — perform focused tasks such as document reading, evidence gathering, and entity matching
  • RAG Semantic Search — retrieves relevant document sections without transmitting data through LLM providers

We focus on secure web application development, actively working to prevent issues such as back-door access, cross-site scripting (XSS), cross-site request forgery (CSRF), SQL injection, clickjacking, and unauthorized cross-origin scripts. SSL/HTTPS is enforced for all communications.

The physical infrastructure is hosted and managed within Amazon's secure data centers utilizing AWS technology. Amazon continually manages risk and undergoes recurring assessments to ensure compliance with industry standards. Amazon's data center operations have been accredited under:

  • ISO 27001
  • SOC 1 and SOC 2/SSAE 16/ISAE 3402
  • PCI Level 1
  • FISMA Moderate
  • Sarbanes-Oxley (SOX)

Additionally, we are committed to compliance with the GDPR. More about GDPR and Heroku compliance: https://devcenter.heroku.com/articles/gdpr

Data Security & Privacy

  • Data Encryption: All data transmitted to and from the platform is encrypted using TLS. Data at rest in PostgreSQL and S3 storage is encrypted using AES-256 encryption provided by AWS.

  • Data Backup: Regular automated backups of the PostgreSQL database are performed via Heroku's continuous protection. Backup integrity is verified periodically to ensure data can be effectively restored.

  • Data Handling: All data collected, stored, processed, and shared by the Mitigate Procurements Platform complies with relevant data protection laws and regulations, including GDPR, to ensure the privacy and protection of user data.

  • Data Classification: The platform primarily handles commercial and technical procurement data (RFP specifications, pricing, certifications, technical proposals). Documents may also contain personal data such as contact persons, signatories, and team member CVs. All uploaded content is treated as confidential by default.

  • Data Usage: Data is only used for purposes explicitly consented to by the users or as required for the provision of the platform's services. Uploaded documents are processed solely for the purpose of procurement analysis.

  • Data Disclosure: User data is not disclosed to third parties without user consent, except where required by law.

  • Data Retention: Customer data, including uploaded documents, analysis results, and conversation logs, is retained for the duration of the active account. Upon account termination or explicit customer request, all associated data is deleted from the platform, including backups within the standard backup rotation cycle.

  • Right to Deletion: Customers may request deletion of specific documents, analysis runs, or their entire account data at any time, in accordance with GDPR Article 17 (Right to Erasure).

AI & LLM Security

The Mitigate Procurements Platform uses artificial intelligence to analyze procurement documents. This section describes the security measures specific to our AI processing pipeline.

Data Flow to AI Providers

  • Anthropic (Claude): Used as the primary AI model for document analysis and reasoning. Document content is transmitted to Anthropic's API for processing. Anthropic's API terms stipulate that customer data submitted via the API is not used for model training. Data is processed transiently and not retained by Anthropic beyond the immediate request.

  • Google (Gemini): Used as an alternative AI model. Google's API terms for enterprise usage provide equivalent no-training guarantees for API-submitted data.

  • LlamaIndex (LlamaParse): Used for parsing uploaded PDF and Word documents into structured text. Documents are transmitted to the LlamaParse cloud API for conversion and are not retained after processing.

  • AgentSet: Used for semantic search (RAG — Retrieval-Augmented Generation). Document content is indexed to enable fast retrieval of relevant sections. AgentSet stores document embeddings in isolated namespaces per customer.

AI Security Controls

  • No Model Training on Customer Data: None of our AI providers use customer-submitted data to train or improve their models. All processing is transient and request-scoped.

  • Prompt Injection Mitigation: The platform implements structured prompt construction with clear separation between system instructions and user-provided document content. Agent tools validate all inputs before execution, reducing the risk of prompt injection attacks.

  • Output Validation: All AI-generated findings (compliance concerns, observations, entity extractions) are validated through structured tool calls with type checking, enum validation, and range verification before being persisted to the database. Critical and major findings undergo an additional automated verification pass.

  • Token and Cost Controls: The analysis engine enforces iteration limits and token budgets to prevent runaway processing. Token usage is tracked per analysis run and per agent type, providing full cost transparency and auditability.

  • Agent Isolation: Sub-agents operate with minimal context — they receive only the specific document sections and instructions needed for their task, not the full conversation history or other customers' data.

  • Conversation Log Security: Full conversation logs between the orchestrator agent and AI providers are stored encrypted and are accessible only to the owning organization's authorized users. Logs can be exported for audit purposes.

Document Processing Security

  • Upload Security: Documents are uploaded via the web interface with content-type validation. Uploaded files are stored in Amazon S3 with server-side encryption (AES-256) and are not publicly accessible.

  • Processing Pipeline: Uploaded documents pass through a controlled pipeline: upload to S3, parsing via LlamaParse (for PDF/Word) or direct reading (for Markdown/spreadsheets), and optional indexing for RAG search. Each stage is executed as an isolated background job.

  • Access Control: Documents are scoped to their parent procurement or bid record. Only users belonging to the owning organization can access documents. There are no shared or public document URLs.

  • Document Deletion: When a document is deleted, the corresponding S3 object, parsed content, and RAG index entries are all removed.

User Access Control

  • Authentication: All users must access the platform using a unique email address and password. Passwords must meet the platform's complexity requirements, including minimum length and the use of both alphanumeric and special characters. Passwords are hashed using bcrypt with an appropriate cost factor and are never stored in plain text.

  • Multi-Factor Authentication: MFA is planned as a future enhancement to provide additional authentication security for user accounts.

  • Session Management: User sessions are managed securely with encrypted, HTTP-only session cookies. Sessions expire after a period of inactivity. Concurrent session control is enforced.

  • Authorization: User access is strictly scoped to their organization. Users can only view and interact with procurements, bids, documents, and analysis results belonging to their organization. Role-based access levels are enforced.

  • Account Review: User accounts are reviewed periodically to ensure that access levels are appropriate and that dormant accounts are disabled.

Access Logging and Monitoring

  • Monitoring Systems: Continuous monitoring is implemented through Heroku's platform metrics and alerting capabilities to detect unusual activities that could indicate security incidents or vulnerabilities.

  • Analysis Audit Trail: Each AI analysis run maintains a complete audit trail including: initiating user, timestamps, documents analyzed, tools invoked, token usage, and all findings generated. This trail is immutable once the analysis is complete.

  • Incident Investigation: In the case of suspected or actual security incidents, detailed investigations are conducted, leveraging access logs, analysis audit trails, and monitoring data to understand the scope and impact.

Multi-Tenant Isolation

  • Organization-Level Isolation: The platform enforces strict multi-tenant isolation at the organization level. Each organization's data (procurements, bids, documents, analysis results, entities) is logically separated in the database and cannot be accessed by users from other organizations.

  • Query-Level Enforcement: All database queries are scoped to the authenticated user's organization, ensuring that data leakage between tenants is prevented at the application layer.

  • RAG Namespace Isolation: Semantic search indexes are maintained in separate namespaces per procurement and per bid, ensuring that RAG queries only return results from the relevant document scope.

  • File Storage Isolation: Uploaded documents in S3 are organized and access-controlled per organization, preventing cross-tenant file access.

Infrastructure Security

  • Shared Responsibility Model: In line with Heroku's shared responsibility model, while Salesforce ensures the security of the Heroku infrastructure, Mitigate Procurements Platform is responsible for securing the application layer and managing access to its Heroku account and resources effectively.

  • Heroku Security Features: The platform utilizes Heroku's security features to enhance its security posture, including automated SSL certificate management, network-level isolation, and customer-configurable features to maintain secure deployments.

  • Environment Variable Security: All sensitive credentials (API keys, database URLs, encryption keys) are stored as Heroku config vars, never committed to source code. Access to production configuration is restricted to authorized personnel.

  • Dependency Security: Application dependencies are regularly audited using automated tools (Bundler Audit for Ruby gems, Brakeman for static analysis, importmap audit for JavaScript dependencies) to detect and remediate known vulnerabilities.

  • Compliance and Certifications:

    • GDPR: The platform adheres to GDPR regulations, ensuring the protection of data for users within the EU. All infrastructure is hosted in the EU region.
    • ISO 27001, 27017, and 27018: Recognition of Salesforce's certifications against these standards ensures adherence to recognized security management best practices and protection of PII.
    • SOC 1, 2, and 3: Salesforce's SOC reports confirm the design and effectiveness of Heroku's controls relevant to the security, availability, and confidentiality of customer data.
    • PCI: Salesforce's Attestation of Compliance as a PCI Level 1 Service Provider is acknowledged. Payment card processing is fully delegated to Stripe (see Payment Security).

Payment Security

  • Stripe Checkout: All payment processing is handled entirely by Stripe via Stripe Checkout. No payment card data is transmitted to, processed by, or stored on the Mitigate Procurements Platform infrastructure.

  • PCI Compliance: PCI DSS compliance is fully delegated to Stripe, a certified PCI Level 1 Service Provider. The platform's PCI scope is limited to redirecting users to the Stripe-hosted checkout page.

  • Webhook Security: Stripe webhook events are verified using Stripe's webhook signature verification to ensure authenticity and prevent replay attacks.

  • Credit System: Customer credits are managed internally after confirmed Stripe payment events. Credit balances are maintained in the platform's database with full transaction audit trails.

Security Audits

  • Regular Audits: Security audits are conducted regularly to assess the platform's compliance with this policy and to identify potential vulnerabilities.

  • Automated Scanning: Static analysis (Brakeman), dependency auditing (Bundler Audit), and JavaScript dependency checks (importmap audit) are run as part of the development and deployment process.

  • Audit Response: Findings from security audits are addressed in a timely manner, with corrective actions implemented to mitigate identified risks.

Third-Party Security

The platform integrates with the following third-party services. Each is assessed for compliance with relevant security standards:

Provider Purpose Data Shared Security Posture
Heroku (Salesforce) Application hosting All application data SOC 1/2/3, ISO 27001, PCI Level 1, GDPR
AWS (S3) Document file storage Uploaded documents SOC 1/2/3, ISO 27001, PCI Level 1
Anthropic AI analysis (Claude) Document content for analysis API no-training policy, SOC 2
Google AI analysis (Gemini) Document content for analysis API no-training policy, ISO 27001, SOC 1/2/3
LlamaIndex Document parsing (LlamaParse) Documents for format conversion Enterprise API terms
AgentSet RAG semantic search Document embeddings Namespace-isolated storage
Stripe Payment processing No card data (redirect only) PCI Level 1, SOC 1/2

  • Data Minimization: Data shared with third-party providers is limited to what is strictly necessary for the service function. Document content is only sent to AI providers during active analysis processing.

  • Contractual Safeguards: Agreements with third-party providers include terms ensuring data protection, confidentiality, and incident notification obligations.

Acceptance and Compliance

All users of the Mitigate Procurements Platform are required to acknowledge and comply with this security policy. Violations of the policy may result in disciplinary action, up to and including termination of access to the platform.

Incident Response

  • Incident Detection: The platform implements monitoring tools and automated alerting to detect potential security incidents, including unauthorized access attempts, unusual API usage patterns, and infrastructure anomalies.

  • Incident Response Plan: In the event of a security incident, a predefined incident response plan is activated. This includes:

    1. Containment — Isolate affected systems to prevent further impact
    2. Eradication — Identify and remove the root cause
    3. Recovery — Restore affected services and verify integrity
    4. Post-Incident Analysis — Document findings and implement preventive measures

  • Data Breach Notification: In the event of a personal data breach, affected customers and relevant supervisory authorities will be notified within 72 hours as required by GDPR Article 33.

  • Communication: Stakeholders are informed about significant security incidents in accordance with the platform's communication protocol. Security incidents can be reported to info@mitigate.dev.

Policy Review and Update

  • Review Cycle: This security policy is reviewed annually or following significant changes to the platform, its hosting environment, or its AI provider integrations.

  • Updates: The policy is updated as necessary to address new security challenges, including evolving AI/LLM security considerations, and to reflect changes in best practices.

Prepared by Mitigate: Edgars Beigarts, Mitigate CTO

Back