Safeguarding artificial intelligence systems against cyber risks

Organisations are adopting Artificial Intelligence (AI) systems apace as they seek to enhance efficiencies, increase optimisation and secure additional benefits. While the opportunities presented by AI are becoming clearer by the day, so too are the associated challenges, including significant new cyber-security risks that AI systems can introduce. When seeking to harness the benefits of AI, it is therefore essential for organisations to adopt a robust approach to safeguarding AI and its benefits. This article looks at how this can be done

07 Aug 2024

Mark Hendry Partner, Head of Cyber Security

Mark Hendry

Authors

Mark Hendry

The evolving field of AI cyber security

The field of AI security is rapidly evolving. Fortunately, researchers, technologists and practitioners across the world are putting significant thought and effort into understanding AI trends. This includes how organisations and cyber criminals are using AI and the ways in which cyber attackers are – or might – attack AI systems. To date, several frameworks and approaches to AI security have been made available including those from MITRE, Open AI, Google, The National Institute of Standards and Technology (NIST), and the European Commission.

Working with you to safeguard your AI

One framework and evolving knowledge base that the Evelyn Partners Cyber Advisory team commonly uses in AI-related work with clients is the MITRE Adversarial Threat Landscape for AI Systems (ATLAS). This incorporates knowledge of real-world AI system attacks collected and analysed by security researchers and ethical hackers who aim to break into, or otherwise subvert, AI systems.

Our use of the framework (and the others mentioned above) enables us to draw on the world’s most advanced research into AI threats and vulnerabilities. We use this knowledge and security expertise to design and implement resilient and fit-for-purpose risk responses, all within the context of our clients’ AI strategies. We:

Consider the attack vectors, tools, techniques, and procedures used by adversaries and select appropriate mitigations to disrupt attack pathways, lessening the likelihood of a security impact occurring on a client’s AI tooling
Model this across the attack pathway from reconnaissance and initial access, through escalation and execution, in order to design and implement safeguards that can disrupt and expunge an attacker’s presence

Speak to Evelyn Partners

As AI adoption continues, adopting a robust AI cyber security strategy alongside your AI strategy is essential. Evelyn Partners can help you. We combine knowledge of emerging AI with solid cyber security expertise to help safeguard our clients’ AI strategies and systems.

If you are adopting, or considering adopting AI, please get in touch with us today.