Intel – Expert Content

Situation

Akamai, one of the most innovative companies in internet technologies, partnered with Intel and IBM to address one of the most vulnerable security areas in computing: data in active memory. While data in storage and data in motion are protected through access control, encryption, firewalls, etc., data in memory remains somewhat unprotected, especially as it typically needs to be decrypted to be processed.

While this issue is of particular importance in financial and health transactions, utilities, voting software, and pretty much everywhere else, the advent of AI brought an additional challenge.

In the past, computer code could be inspected and a malicious insertion or injection could be tracked as programming languages had tens or at the most a couple of hundred reserved words in the programming vocabulary (e.g., IF-THEN-ELSE, TRY-CATCH-THROW, SELECT-FROM-WHERE, etc.).

With the emergence of the Large Language Models (LLMs) used in AI engines, instructions can be given using tens of thousands (or more) words. A malicious AI trainer can insert instructions to take over control or generate damage when a certain sequence of words is presented to the interpreter, Manchurian Candidate style. It is practically impossible to back-engineer AI training and therefore such an intrusion would be extremely difficult to track.

Transiris was tasked to explain in an awareness video this quite esoteric issue of Confidential Computing and the Intel Tiber Trust Services solution developed by Intel, supported by IBM, and managed by Akamai dedicated applications that address the problem.

In addressing these challenges, the campaign also explored advanced cryptographic protocols, secure enclaves, and hardware root of trust mechanisms to reinforce data protection at the processor level.

This approach leveraged virtualization and containerization technologies to illustrate how dynamic security measures can be integrated into modern computing environments.