Skip to main content
BlogCybersecurity
Cybersecurity

New AI Cyber Threats Need New Defenses

Traditional cyber tools stop attacks that have a known signature. AI-assisted autonomous offensive frameworks have no signature, only behaviors that change, so enterprises need new defenses.

ST
Sentrix Team
Product Operations
November 19, 20263 min read

AI-assisted autonomous offensive frameworks (such as ARTEX) require enterprises to build new defenses. That’s the best summary of this entire article. Simply stated, traditional cyber tools work well against traditional attacks that have a well defined signature. However, new AI-assisted offensive frameworks often have no signature.

There’s a lot of hype surrounding AI-assisted threats. Is the solution to patch everything? In theory, but in practice patching everything may increase business downtime from hastily deployed patches. Are virtual patches (aka compensating controls) a cure all? Not really, as a virtual patch may block one attack vector only for the AI to find another path to the vuln. And of course, cyber best practices are necessary, such as least privileges, network segmentation and so on. All of these traditional techniques are as critical as ever, but are no longer sufficient.

AI-assisted autonomous offensive frameworks are not changing the attack surface, but they are changing a corporation’s cyber exposure.

Let’s cut straight to one critical new consideration when defending against these new AI-assisted attacks:

Traditional threats have a knowable signature. And traditional cyber tools can protect your corporation when they identify these signatures on your systems.

However, AI-assisted threats do not have a knowable signature, so traditional tools are not effective against these new threats. Instead, these new threats have behaviors. Making matters more complicated, these new AI threats have dynamically changing behaviors so you are looking for indicators of an attack.

An example will help. An AI-assisted attack may perform the following steps, all of which your actual users may do in the normal course of business.

  1. Download an API specification using a normal GET request.
  2. Perform a password reset. Perhaps trying a few variations of a username which users may do if they forgot their username.
  3. Read a client secret from a file. It’s totally normal for users to read files.
  4. Make a series of valid OAuth calls to validate the client secrets. Each of these are independent and valid OAuth calls.
  5. Use a validated secret to call a protected API. No problem here as the credentials are real and valid.

Where cyber defenders earn their keep is in finding the oddities and anomalies in these behaviors. Is a valid credential used on a new IP too quickly to be human? It’s subtle, but an indicator of compromise. Are tools being chained together and combined with high frequency requests? Is a valid credential calling an API out of cycle? For example, is a credential + API used for monthly rollups being used in the middle of the month? The list of behaviors that may indicate compromise is substantial.

It’s easy to see why these new attacks are a challenge. Making things more complicated, a skilled threat actor may prompt their AI-assisted offensive framework to goal seek, so when defenders block a behavior (ex. too many tools called in rapid succession), then the offensive framework may change its behavior in response to your defense.

We’re in a new era of cybersecurity. And new defensive cyber tools are needed to defend against AI-assisted autonomous offensive frameworks.

Share this article
ST

Sentrix Team

Product Operations

Sentrix Team is a contributor at Sentrix.ai, focusing on cybersecurity topics and multi-agent orchestration systems.

Related articles

Find out which of your controls still act in time.

A fixed-scope assessment of your posture against autonomous offensive frameworks.

Request the assessment