DATA BREACHES

Mayank Upadhyay Chief Security, Trust Officer and Vice President of Engineering Snowflake

The Cyber LLM has entered the chat The pace of AI development means that defensive security strategies can become obsolete in a matter of months.

Mayank shares how rapid hardware and architectural breakthroughs are radically altering what these models can process:

“ Every three to six months the LLM world takes such big strides. There’ s been major advances both in GPU technology as well as in concepts like being able to keep more context, having more memory. So these LLMs can basically keep a lot more state in their brain and they can connect the dots between things which might have normally felt like separate things.”

In a cybersecurity context, this enhanced brainpower allows an AI to map out complex attack vectors.“ They may find a vulnerability here and a vulnerability there, and they can figure out how to chain these things together and very soon find a path from one place to another,” Mayank explains.

While commercial models are built with safety restrictions, a new breed of unrestricted AI – the Cyber LLM – is emerging. Mayank continues:“ Normally, when you have these kinds of capabilities, you also put guardrails on them and you say‘ thou shall not do anything evil’. But a cyber LLM is basically one where you’ ve removed those constraints. the customer is going to point you at their own source code and their own environment and they want you to find stuff. So, go at it.”

cybermagazine. com 103