New Cybersecurity Guidance: Securing the Data Powering AI Systems

Artificial intelligence is transforming industries and enabling ground-breaking capabilities. As companies increasingly rely on AI to power critical systems and inform high-stakes decisions, ensuring the security of the data fueling AI is paramount.

A new Cybersecurity Information Sheet (CSI), jointly authored by the National Security Association (NSA), Cybersecurity and Infrastructure Agency (CISA), and cybersecurity authorities from key allies, provides essential guidance on securing data used to train and operate AI systems. The document outlines critical best practices spanning the entire AI lifecycle.

Foundational measures include sourcing data from authoritative providers, safeguarding data integrity, authenticating revisions with digital signatures, leveraging trusted infrastructure, applying robust encryption and access controls, and conducting ongoing risk assessments. Organizations must incorporate these practices from the initial planning stages through development, testing, deployment and monitoring.

The CSI takes a deep dive into three significant risk areas: data supply chain threats, maliciously modified data, and data drift. Compromised data can enter AI systems through vulnerabilities in web-scale datasets. Malicious actors can poison data to manipulate model behavior. Statistical biases and inaccurate information degrade performance. And data distributions inevitably shift in production environments.

For each risk area, the CSI provides clear mitigation strategies. From anomaly detection to collaborative learning, metadata validation to deduplication, and continuous monitoring to dynamic updates - a multi-layered approach is essential. Proactive data security is a necessity, not an option.

As AI grows increasingly ubiquitous and impactful, organizations cannot afford to neglect the integrity of their AI data. Those who adopt rigorous data security measures will be well-positioned to reap AI's immense benefits while safeguarding against potentially disastrous harms. Prioritizing AI data security is a strategic imperative for any organization seeking to responsibly leverage this transformative technology.

Access the AI Data Security information sheet.