The One Question Boards Should Be Asking Their Companies: “How Are You Securing Your AI?”

The AI revolution is upon us. Venture capital has poured billions into companies promising to unlock the transformative potential of artificial intelligence. But amidst the excitement, a critical question often gets overlooked: “With cyberattacks being so widespread, how can you safely implement AI?”

While investors are understandably focused on the innovative applications of AI – from self-driving cars to revolutionizing healthcare to optimizing supply chains – they must not lose sight of the critical importance of security. As AI systems become increasingly sophisticated and integrated into our lives, evolving from simple responders to autonomous agents capable of executing real-world actions, the risks associated with data breaches, malicious exploitation, and unauthorized access become exponentially greater.

Consider these scenarios:

• A healthcare AI company: How can AI be given appropriate context about medical data in a secure and federated way?
• A financial technology company: How is the company safeguarding customer financial data, preventing fraud, and ensuring that AI-powered trading algorithms operate ethically and responsibly? A breach of financial data could lead to significant financial losses for customers and damage to the company’s reputation.
• A global logistics company: How is the company ensuring the security of its AI-powered supply chain management system, preventing hackers from manipulating inventory levels, disrupting shipments, or stealing sensitive data like supplier contracts and customer information? A breach in the supply chain could have severe consequences, from production delays and financial losses to reputational damage and even national security implications.

These are just a few examples of the critical security challenges facing AI companies across various sectors.

Beyond “How are you using AI?”

Boards must shift their focus beyond simply asking “How are you using AI?” to a more fundamental question: “How are you securing this?”

This requires a deep dive into the company’s security posture, including:

• Access Control
  • Which AI agent can access which data? For example, in a healthcare setting, can an AI agent designed for appointment scheduling access sensitive medical records?
  • Which person can access which AI agent? Can a junior employee access and modify the parameters of a critical AI system used for medical diagnosis?
  • Which AI agent can access which models? Can an AI agent designed for customer service access and utilize machine learning models developed for fraud detection?
  • Which AI agents can access which other agents? How are interactions between different AI systems within the company controlled and secured?
  • Fine-grained permissioning: Can access be easily granted, revoked, and modified based on user roles, context, and time?
• Data security
  • Encryption: Are all data transmissions and storage encrypted using robust, industry-standard algorithms?
  • Data privacy: What measures are in place to protect sensitive data, such as patient information, financial data, and intellectual property from unauthorized access and misuse?
• System security
  • Authentication and authorization: How are identities verified, and how are access controls implemented to prevent unauthorized access to sensitive systems and data?
    Vulnerability management: How many attack surfaces are visible to outsiders? How can you reduce or even eliminate them?

The Limitations of Perimeter Security and Microsegmentation

Perimeter security models, with their focus on firewalls and network segmentation, are ill-suited for the complex and dynamic environment of modern AI systems, especially in sectors like healthcare, finance, and supply chain management where interactions with external entities are frequent and critical.

• Complexity: AI systems often involve intricate networks of interconnected devices, applications, data stores, and cloud services, making perimeter security difficult (if not impossible) to implement and manage effectively.
• Attack Surfaces: Open ports and vulnerabilities in interconnected systems create numerous entry points for attackers, rendering perimeter defenses increasingly ineffective.
• Zero Trust Challenges: The principle of “zero trust”—that no person, entity, or device is inherently trusted and should only have access to the minimum amount of data required to perform its role—is difficult to implement within a perimeter-based security model, especially when dealing with a constantly evolving network of partners, suppliers, and customers.

The Need for Organic Network Security

To address these challenges, a new approach to security is needed—one that focuses on identity and trust.

• Simplified Access Control: An identity-based security model simplifies access control with fine-grained permissioning and flexible granularity, allowing users and AI agents to have precise control over data access and interactions.
• Built-in Cryptographic Authentication: Strong cryptographic authentication mechanisms ensure that only authorized entities, whether human users, AI agents, or external systems, can access and interact with the rest of the system safely.
• Invisible to Adversaries: An organic network security effectively eliminates attack surfaces by moving them to points that have nothing of value and do not allow direct access to the systems that do. Thus attackers cannot gain unauthorized access to anything of value.
• Zero Trust Architecture: An identity-based approach aligns with the principles of zero trust, where trust is not implicitly granted but must always be verified.
• Contextual Decision Making: Enables AI agents to provide different responses to the same request depending on the identity and authorization level of the requesting entity, ensuring that only authorized parties have access to sensitive information.

Investing in the Future of Trust

Investing in AI companies requires not just an assessment of their technological prowess but also a rigorous evaluation of their security posture. Companies that prioritize security and build trust with their customers will be best positioned for long-term success.

By asking the right questions and prioritizing security from the outset, investors can help ensure that the AI revolution unfolds in a safe, secure, and ethical manner.