AI & Data Companies

AI and data companies face risks that standard control sets often miss: prompt injection, sensitive information disclosure, model and data poisoning, supply-chain vulnerabilities, and excessive agent autonomy.

Our approach: Standard NY SHIELD Act readiness first. AI and data-specific hardening second. The advisory modules below are optional enhancements on top of mandatory controls.

Standard Controls vs. AI/Data Enhancements

Standard NY SHIELD Act Readiness

Mandatory controls required for compliance:

  • Logical access and privileged access
  • Change management
  • Incident response
  • Risk management
  • Vendor management
  • Backup and availability
  • Logging and monitoring
  • Confidentiality and privacy (where applicable)

View all control domains →

AI/Data Advisory Enhancements

Optional modules justified by AI-risk frameworks:

  • Data lineage and training data governance
  • Prompt/response telemetry
  • RAG and retrieval governance
  • Model/provider vendor review
  • Agent approval gates
  • AI-assisted SDLC controls
  • Warehouse and analytics governance

Advisory Modules

Each module adds specific controls and documentation practices to address risks unique to AI and data-intensive products.

AI-Assisted SDLC

NY SHIELD Act technical safeguards require assessing risks in software design (§ 899-bb(2)(b)(ii)(A)), which extends to development workflows augmented by AI code generation and automated development tools.

What This Module Adds

  • Security review of AI-generated code for unintended collection or processing of private information as defined under § 899-aa(1)(b)
  • Automated scanning of AI-suggested features for new data flows involving NY resident private information
  • Risk assessment of AI coding tool access to development environments containing private information
  • Documentation of AI tool usage in security-critical code paths to support safeguard sufficiency evaluations

Human Review & Agent Gates

Administrative safeguards require training and managing employees in security practices (§ 899-bb(2)(b)(i)(C)); AI agents handling private information need equivalent oversight to ensure safeguard sufficiency.

What This Module Adds

  • Human approval gates for AI agent actions that access, modify, or transmit private information
  • Audit logging of AI agent decisions involving NY resident data to support breach investigation and notification timelines
  • Role-based access controls for AI agents consistent with the principle of least privilege for private information access
  • Escalation procedures when AI agents encounter data classified as private information under the SHIELD Act's expanded definition

Model Provider Vendor Risk

Section 899-bb(2)(b)(i)(E) requires selecting service providers capable of maintaining appropriate safeguards and requiring those safeguards by contract — a mandate that applies directly to AI model providers processing private information.

What This Module Adds

  • Vendor security assessments for AI model providers evaluating their ability to maintain safeguards for private information
  • Contractual provisions requiring AI model providers to implement administrative, technical, and physical safeguards for NY resident data
  • Data processing agreements specifying prohibitions on using private information for model training without explicit authorization
  • Ongoing monitoring of AI model provider security practices and incident notification capabilities
  • Exit and data deletion provisions ensuring private information is disposed of when the vendor relationship ends

Prompt & Response Logging

Technical safeguards require detecting, preventing, and responding to attacks or system failures (§ 899-bb(2)(b)(ii)(C)); logging AI interactions provides detection evidence and supports breach investigation timelines.

What This Module Adds

  • Structured logging of AI prompts and responses that may contain or reference private information
  • Automated detection of private information categories (biometrics, email-password pairs, financial account data) in AI interaction logs
  • Retention policies for AI interaction logs aligned with the organization's data retention schedule and breach investigation needs
  • Access controls on AI interaction logs to prevent unauthorized exposure of private information captured in prompts or responses

RAG & Vector Store Controls

Vector stores containing private information are subject to all three safeguard categories under § 899-bb(2)(b) — administrative, technical, and physical — as they constitute systems that store and process NY resident data.

What This Module Adds

  • Data classification review before ingesting documents into vector stores to identify private information subject to SHIELD Act protections
  • Access controls on vector store queries to prevent unauthorized retrieval of embedded private information
  • Encryption of vector embeddings derived from documents containing private information
  • Disposal procedures for vector store entries when source documents containing private information reach end of retention
  • Vendor assessment for hosted vector database providers under § 899-bb(2)(b)(i)(E) service provider requirements

Training & Inference Data Governance

Using private information for model training creates risks to the security, confidentiality, and integrity of that information that must be assessed under the risk assessment safeguard requirement of § 899-bb(2)(b).

What This Module Adds

  • Risk assessment of model training datasets for inclusion of NY resident private information under the SHIELD Act's expanded definition
  • Data minimization controls to limit private information exposure during model training and fine-tuning
  • Inference pipeline controls to prevent model outputs from leaking memorized private information
  • Documentation of training data provenance to support breach investigation if a model is found to have memorized private information

Warehouse & Analytics Governance

Data warehouses holding NY resident private information are in scope for all three safeguard categories under § 899-bb(2)(b), as they constitute centralized systems for storing, processing, and analyzing private information at scale.

What This Module Adds

  • Column-level classification of warehouse tables to identify fields containing private information under the SHIELD Act's expanded definition
  • Query access controls and audit logging for warehouse tables containing NY resident private information
  • Data masking and anonymization policies for analytics workloads that do not require access to raw private information
  • Retention and disposal automation for warehouse records containing private information that has exceeded its business-purpose retention period
  • Vendor assessment for cloud data warehouse providers under the service provider safeguard requirement

Need AI-Specific Readiness Support?

We help AI and data companies build a control environment that satisfies enterprise buyers and addresses the unique risks of AI products.

Get in Touch