With AI upending the nature of global enterprise, security leaders must take a holistic look at how their cyber defenses might be impacted.
Enterprises must understand the dangers across software and hardware. They must grasp that AI’s emergence means new measures are required to protect endpoints carrying key data.
North American security leaders agree with Foundry research showing 61% either have endpoint security or are in the process of upgrading it. [1]
Speaking on a CIO.com webcast, JR ‘JRB’ Balaji, Head of Software Product Management for the AMD Client Business Unit, says that while the fundamentals of a good security policy – such as architecture and hygiene – still apply in the age of AI, the new technology represents a paradigm shift.
He adds: “Organizations need to be thinking about both security for AI and AI for security.”
The last ten years have seen IT and security leaders facing an increasingly complex and challenging threat landscape. Cyber criminals and even nation states are leveraging toolkits to their best advantage and growing more professional in their approach.
They’re using AI and, by utilizing large language models, are developing optimized or polymorphic malware that is resistant to detection or creating more sophisticated phishing attacks.
What’s more, AI has created new targets, such as hackers attacking enterprise AI services with malicious code, prompt injections and corrupt training data, or attempting to infiltrate systems to steal and reconstruct AI models.
Even end users can raise risk levels by entrusting confidential corporate and customer data to open-source or public AI tools.
Two key security goals
For JRB protecting the business and its data comes down to two things: how to minimize the attack surface and, second, how to be resilient against attacks when they happen.
Resilience requires a mix of proactive and reactive approaches, with proactive resilience covering the protections and countermeasures put in place to defend against attack, and reactive resilience covering how organizations can get back up and running in the event of an incident.
JRB suggests that “organizations should lean in as much as possible on proactive cyber-resilience, where the foundation starts with the silicon and hardware and extends all the way up to the operating system and the data layers.”
That means multiple layers of defense, beginning with features such as AMD Secure Boot [2], which help ensure that only trusted firmware and bootloaders, verified by digital signatures, can be loaded during system startup.
It also means encrypting data both at rest, in transit and while it’s being processed, so that it’s protected in the event of an intrusion.
Crucially, enterprises need to take advantage of new processor technologies designed to accelerate or secure local AI workloads, so that more can be done with sensitive data on the device, rather than in public applications on the cloud, where it may leak out.
The Neural Processing Units (NPUs) in the latest processors can help by enabling AI-driven security capabilities as well as on-device AI processing.
Meanwhile, support for confidential computing can help enterprises protect their data and their models from adversaries looking to modify code or tamper with the training data.
Watch JRB’s full interview in this CIO.com webcast.
[1] Foundry, Security Priorities Study 2024
[2] An OEM who has enabled the AMD Platform Secure Boot feature grants permission for their cryptographically signed BIOS code to run only on their platforms using an AMD Platform Secure Boot enabled motherboard. One-time-programmable fuses in the processor bind the processor to the OEM’s firmware code signing key. From that point on, that processor can only be used with motherboards that use the same code signing key. GD-192.