The Cybersecurity Infrastructure & Security Agency (CISA) recently published AI Data Security: Best Practices for Securing Data Used to Train & Operate AI Systems in collaboration with the National Security Agency's Artificial Intelligence Security Center (AISC), the Federal Bureau of Investigation (FBI), the Australian Signals Directorate's Australian Cyber Security Centre (ASD's ACSC), New Zealand's Government Communications Security Bureau's National Cyber Security Centre (NCSC-NZ), and the United Kingdom's National Cyber Security Centre (NCSC-UK).
The document provides guidance on securing data used in artificial intelligence and machine learning systems, highlighting data security's importance in ensuring accuracy and integrity of AI outcomes while outlining potential risks from data integrity issues across AI development and deployment stages.
According to the guidelines, six essential components comprise a comprehensive AI data security program. Continuous encryption—protecting data in transit, at rest, and in use—forms the foundation, playing a critical role throughout the AI lifecycle.
Continuous Data and Model Protection Requires Continuous Encryption
Encryption is fundamentally necessary. Organizations share sensitive corporate data with AI models to enhance decision-making, automate workflows, and unlock operational gains. However, current AI tools often lack robust security capabilities protecting against malicious access, data breaches, and adversarial attacks like model poisoning. These vulnerabilities enable hackers to manipulate AI outputs, compromise decision-making, and expose organizations to regulatory penalties for violating data privacy mandates including GDPR and HIPAA.
AI systems process massive amounts of sensitive data requiring protection at every stage. However, CISA's encryption guidelines lack specificity regarding processing: "Adopt advanced encryption protocols proportional to organizational data protection levels, including securing data at rest, in transit, and during processing. AES-256 encryption represents the de facto industry standard and resists quantum computing threats."
While advanced encryption ensuring data protection across all states proves essential, the guidance emphasizing AES-256 as the industry standard is incomplete. AES-256, though potentially quantum-resistant, doesn't natively support computation on encrypted data. As a symmetric-key encryption standard designed for confidentiality rather than ciphertext operations, data must be decrypted for analysis, creating protection gaps.
DataKrypto FHEnom for AI™ Sets the Standard for AI and Data Security
Continuous encryption enables models to be trained and analyzed on encrypted data, protecting both the model and inputs from tampering or exposure. Specifically, DataKrypto's FHEnom for AI™ uses fully homomorphic encryption (FHE) and Trusted Execution Environments (TEEs) to create a zero-knowledge AI framework. This dual-layer security protects AI models from theft and data leakage by keeping data encrypted throughout its lifecycle, including training, inference, and deployment, while ensuring models only function within TEE.
FHE enables direct computation on encrypted embeddings, keeping both model weights and user data protected in ciphertext throughout processing. TEEs provide hardware-enforced isolation for secure tokenization and output within a cryptographically verified enclave. Consequently, queries and AI models remain blind to each other throughout the query, analysis, and output process.
If an AI model is stolen, it becomes useless since it only functions within the TEE. The encryption key, existing solely within the TEE, protects data by ensuring only users possessing the key can view query-generated results.
DataKrypto's approach promises transforming AI data security for several reasons:
- Encrypting both the model and training data
- Allowing models to be trained on or queried against encrypted data so breaches yield useless stolen assets
- Restricting access and data interpretation to authorized users with proper encryption keys
While CISA's collaborative efforts establish necessary AI security guidelines, protecting data and AI models requires more advanced capabilities than suggested. DataKrypto provides solutions ensuring continuous data and model security, allowing organizations to leverage AI's benefits without fearing malicious activities.