Cybersecurity Takeaways from AI's Struggle with Hands

By Grant McDonald · April 8, 2024

Recently, I attempted to leverage an online tool to create AI-generated artwork. As an artist I have mixed feelings about the use of AI in creating art but as a cybersecurity veteran I see other implications from my experience.

I set out to create a unique piece of AI-generated art. I went through several iterations attempting to train a model to generate artwork based on a short story, visuals, artistic styles and cinematography prompts. The results were pretty interesting but one thing I noticed very early on - why can’t AI get hands right?

My story about a marketer concerned about cyberthreats who seeks help from a galaxy far, far away is entertaining but look closely and you will notice a few things. Try as I might, over and over hands came out either disfigured, incorrect, or forged with other parts of the image. Backgrounds were also challenging as was getting characters to look consistent. I ultimately revised the story to fit the images but even then things are not quite right. That’s not the way you hold an “elegant weapon for a more civilized age” for one!

Despite its impressive capabilities, the AI occasionally struggled with certain aspects, producing inconsistencies that betrayed its synthetic nature. These errors served as a reminder that while AI has made remarkable strides in replicating human creativity, it still falls short of perfection.

AI-powered security solutions are similar in that they are not immune to imperfections. While AI has revolutionized threat detection and response, it cannot yet fully automate security operations. Like the AI-generated artworks, security algorithms may encounter errors or overlook subtle indicators of malicious activity, leaving organizations vulnerable to cyber threats.

This is why AI coupled with Extended Detection and Response (XDR) is increasingly important. AI is a “garbage in, garbage out” technology that is dependent on getting the right data. When XDR provides data from multiple security layers and tools to AI and machine learning, models can be better informed or refined.

XDR, AI and Machine Learning still require human involvement. Much like my AI generated art above, someone may need to examine if a behavior is truly malicious and determine the proper response. As weaponized AI increases and threat actors focus on targeting GenAI, teams will also have to refine or train models on new behaviors. That doesn’t mean there isn’t a wealth of benefits from XDR and AI today. XDR powered by AI can take a significant chunk of the guess-work and alert fatigue out of the picture. It can make recommendations of what to investigate or remediation steps to consider. GenAI can make common investigation and response accessible to more of your security team no matter what level of experience they have currently.

At Trellix we have a long history of Machine Learning and AI. Since 2010, we have continually added analytics, machine learning and AI to our security tools. Our latest development is the addition of GenAI for faster incident detection and response. Just as XDR augments machine learning and analytics capabilities with multi-vector data; our GenAI capabilities complement our XDR platform by raising context, automatically investigating and guiding teams through responses faster.

With this powerful combination of XDR, AI and GenAI you can lean into use cases such as:

• Protecting against data leakage when employees are using GenAI
• Using AI driven analytics for greater data accuracy and predictive analysis
• Triaging processes that GenAI can handle with speed and scale
• Automating the creation of security content

You can learn more about the applications, use cases and how Trellix can help you leverage AI in your security operations in this post from our CISO, Harold Rivas.