6-RFP-26-00210 Exhibit 3 - AI Security Questionnaire.xlsx
XLSX spreadsheet 40 KB Posted
- Attached to
- Payment and Program Management Software Platform
- Solicitation number
- RFP-26-00210
- Issued by
- Hillsborough County, Florida
About this file
This is an AI Security Questionnaire for the Hillsborough County Board of County Commissioners' Payment and Program Management Software Platform procurement (RFP-26-00210), a cloud-based solution with an estimated project value of $555,000 over a three-year contract term. The questionnaire requires vendors to complete vendor identification information and respond comprehensively to security and compliance questions across eight categories: Document Requests, Asset Management, Governance, Life Cycle Management, Data and Model Integrity, AI Security, Privacy, and AI Incident Management. Vendors must provide supporting documentation including audit reports (SOC 2), penetration test and vulnerability assessment reports, AI architecture diagrams, and Model or Transparency Cards. The questionnaire addresses critical security controls, information classification schemes, data flow mapping, OWASP compliance, and NIST AI Risk Management Framework alignment.
The questionnaire establishes mandatory security and operational requirements for vendors bidding on this Florida county procurement, with specific focus on AI-specific security controls, third-party risk management, and incident response protocols. Vendors must document their approach to data governance, model integrity, privacy compliance, and incident management, including controls for adversarial data poisoning, prompt injection mitigation, model versioning, and 24/7 security monitoring. Response options are standardized as "Yes," "No," "Partial (Please explain)," "Not applicable (Please explain)," or "Attached/Not attached (Please explain)" for document requests. The questionnaire requires vendors to address Hillsborough County data handling, FedRAMP certification levels where applicable, third-party security agreements, and forensic investigation capabilities, ensuring comprehensive vendor security vetting aligned with government and industry security standards.
View the file
Other files
On GovTribe
Work with this file
- Download the original file
- Contacts named in this file
- Similar government files
- Ask GovTribe AI about this file
Text version
AI Questionnaire
| AI Vendor Security Questionnaire | ||
| Vendor to complete vendor identification section and then respond to all questions. If documents are requested, please use "Insert object" to attach the document directly into the Vendor Comments cell. | ||
| Vendor Name: | e.g. ACME Inc. | |
| Completed By: | Name of person responding | |
| Date Completed: |
| # | Question | Vendor Response | Vendor Comments |
| 1 | Document Requests | ||
| 1.2 | Please attach a copy of any relevant audit reports that cover information security controls (e.g. SOC 2). | Attached | |
| 1.3 | Please attach a copy of your latest penetration test and/or vulnerability assessment report. | Attached | |
| 1.4 | Please provide AI architecture diagram illustrating end-to-end AI workflows, including data ingestion, processing, model interaction, and output layers, with emphasis on data flow, governance, and security controls. | Attached | |
| 1.5 | Please attach copies of Model Cards or Transparency Cards. If you don't have them, explain why? | Attached | |
| 2 | Asset Management | ||
| 2.1 | Do you maintain an inventory of all hardware and software,AI/ML, and data(training,operational or other assets, including ownership? | Partial (Please explain) | |
| 2.2 | Do you have an information classification scheme and process designed to ensure that information is protected according to its confidentiality requirements? | Yes | |
| 2.3 | Do you maintain an inventory or map of data flows between both internal and external information systems? | No | |
| 2.4 | Do you classify each data type (training, operational or other data) used by the AI systems | Yes | |
| 2.5 | Do you adhere to OWASP guidelines for application security, including the OWASP Top 10 for AI/LLM applications? | ||
| 2.6 | Are AI-specific gaps identified and documented? | ||
| 2.7 | Are parameter values tailored for AI system risks? | ||
| 3 | Governance | ||
| 3.1 | Have you documented your AI risk management approach in alignment with NIST AI RMF core functions (Govern, Map, Measure, Manage)? | Yes | |
| 3.2 | Can you provide evidence of controls for privacy, security, and resilience as per NIST AI RMF guidance? | Not applicable (Please explain) | |
| 3.3 | Do you have a process for reviewing your information security policy? | Partial (Please explain) | |
| 3.4 | Do you conduct regular risk assessments for fairness, transparency, and accountability as outlined in NIST AI RMF? | Partial (Please explain) | |
| 3.5 | Have you designated an individual, who is at least at a manager level, who is responsible for information security activities? | No | |
| 3.6 | Do you have a process designed to monitor changes to regulations and ensure compliance with relevant security requirements? | Yes | |
| 3.8 | Are controls in place to detect adversarial data poisoning or manipulation? | ||
| 3.9 | Is model versioning documented with traceability of updates? | ||
| 3.10 | Are availability requirements defined for the AI system? | ||
| 3.11 | Are backup/restore procedures for model artifacts implemented and tested? | Partial (Please explain) | |
| 3.12 | Is the AI system explainable and transparent to stakeholders? | ||
| 3.13 | Are safeguards in place for bias, fairness, and harmful output? | ||
| 3.14 | Describe processes for bias testing and mitigation in AI models | ||
| 3.15 | Are adverse events or misuse incidents tracked? | ||
| 3.16 | Are mechanisms in place to detect model drift or degradation? | ||
| 4 | Life Cycle Management | ||
| 4.1 | Is an AI system lifecycle defined from design to decommissioning? | ||
| 4.2 | Has AI ownership and accountability been defined across the lifecycle? | ||
| 4.3 | Have you implemented controls aligned with the NIST AI Risk Management Framework (AI RMF) for managing AI-related risks across the lifecycle? | ||
| 4.4 | Are prompt restrictions or output filtering mechanisms implemented? | ||
| 4.5 | For agent systems, is supervision and oversight defined? | ||
| 4.6 | Are periodic performance and security reviews conducted? | ||
| 4.7 | Are AI system logs and audit trails maintained? | ||
| 5 | Data and Model Integrity | ||
| 5.1 | Do any AI components train on Hillsborough County data? If yes, describe isolation and consent controls.” | ||
| 5.2 | Where is AI training performed (region, cloud provider)? | ||
| 5.4 | Are training data sources validated for accuracy, integrity, and provenance? | ||
| 5.5 | Are controls in place to detect adversarial data poisoning or manipulation? | ||
| 5.6 | Do you mitigate prompt injection and insecure output handling in your AI applications (OWASP LLM01 & LLM02)? Please explain | ||
| 5.7 | Is model versioning documented with traceability of updates? | ||
| 6 | AI Security | ||
| 6.1 | Does the AI system fall into a high‑risk category | ||
| 6.2 | Does your information security policies include content on AI security? | ||
| 6.3 | Are security measures implemented to safeguard AI systems from external threats, such as cyberattacks or unauthorized access? | ||
| 6.4 | Does your AI system have a FedRAMP instance/environment with the necessary security standards, that provides a framework for agencies/local governments to determine if the cloud service falls within the required FedRAMP level? What impact level is it certified? Level(1-6) | ||
| 6.5 | Are you outsourcing any AI processing to a 3rd party? If yes, do you have an agreement that holds them accountable for the data protection? | ||
| 6.6 | Does your incident response plan include AI incidents in their scope? | ||
| 6.7 | Do you perform periodic testing on your AI system/components that includes but is not limited to penetration testing and red-teaming? | ||
| 6.8 | Do you implement input validation and sanitization measures to ensure user input adheres to defined limits and filters out malicious content? | ||
| 6.9 | Do you have appropriate authentication practices in place, such as OAuth 2.0, to implement effective authorization and access control to AI systems? | ||
| 6.10 | Do you enforce privilege control on LLM access to backend systems and follow the principle of least privilege by restricting the LLM to the minimum level of access required for its intended use? | ||
| 6.13 | Do you enforce containerization on all mobile devices that may contain organizational data, including email, whether those devices are owned by the organization or by employees? | ||
| 6.14 | Do you have the capability of deleting all organizational data from mobile devices, whether owned by the organization or by employees, in the event that the device is lost or stolen? | ||
| 6.15 | Does the AI system process sensitive data and is it properly protected? If yes, how do you protect against sensitive information disclosure and model theft (OWASP LLM06 & LLM10)? | ||
| 6.16 | Are outputs reviewed for unintended disclosure of sensitive information? | ||
| 6.17 | Are model weights and configurations encrypted at rest and in transit? | ||
| 6.18 | Are third-party AI components subject to contractual cybersecurity requirements? | ||
| 6.19 | Are third-party model dependencies assessed for vulnerabilities? | ||
| 6.20 | Are vendor updates tested for security impact? | ||
| 6.22 | Is access to model artifacts restricted by authentication and authorization? | ||
| 6.23 | Is the AI deployment environment secured (patching, segmentation, monitoring)? | ||
| 7 | Privacy | ||
| 7.1 | Do you have a data retention policy and process that is designed to meet relevant privacy regulations? | ||
| 7.2 | Do you maintain an inventory and mapping of where all personal data is stored that includes cross-border data flows? | ||
| 8 | AI Incident Management | ||
| 8.1 | Are security configuration baselines defined and implemented for all endpoints and network devices? | ||
| 8.2 | Do you segregate your network into zones based on trust levels and control the flow of traffic between zones? | ||
| 8.3 | Does your incident response plan include AI-specific threats (prompt injection, model compromise)? Provide playbook excerpts. | ||
| 8.4 | Are all AI security incidents recorded, classified, and tracked? | ||
| 8.5 | Are forensic investigations conducted as part of incident response? | ||
| 8.6 | Have you deployed automated tools to collect, correlate, and analyze security event logs from multiple sources for anomalies? | ||
| 8.7 | Are security alerts monitored 24/7? | ||
| 8.8 | Are interactions via APIs or prompts monitored for misuse? |
3. Lists (Hidden)
| Data Entry Lists |
| This tab controls the lists that are used for the data entry tab. Hide this tab when not in use. |
| Valid responses for document request questions |
| Attached |
| Not attached (Please explain) |
| Not applicable |
| Valid responses for all other questions |
| No |
| Partial (Please explain) |
| Yes |
| Not applicable (Please explain) |
File details come from the government source that posted it. Updated .