The realities of AI in cybersecurity: catastrophic forgetting

Issue 2 2021 Information Security

There is a lot of hype about the use of artificial intelligence (AI) in cybersecurity. The truth is that the role and potential of AI in security is still evolving and often requires experimentation and evaluation.

Malware detection is the cornerstone of IT security and AI is the only approach capable of learning patterns from millions of new malware samples within a matter of days.

But there’s a catch: should the model keep all malware samples forever for optimum detection but slower learning and updates; or go for selective fine-tuning that enables the model to better keep up with the rate of change of malware, but runs the risk of forgetting older patterns (known as catastrophic forgetting)?

Retraining the whole model takes about one week. A good fine-tuning model should take about one hour to update.

SophosAI wanted to see if it was possible to have a fine-tuning model that could keep up with the evolving threat landscape, learn new patterns but still remember older ones, while minimising the impact on performance. Researcher Hillary Sanders evaluated a number of update options and has detailed her findings in the Sophos AI blog (https://ai.sophos.com/blog/).

The detection dilemma

Keeping detection capabilities up to date is a constant battle. With every step we take towards defending against a malicious attack, adversaries are already developing new ways to get round it, releasing updates with different code or techniques. The result is that hundreds of thousands of new malware samples appear every day.

Detection is made even harder by the fact that the latest-and-greatest malware is rarely completely ‘new’. Instead, it is more likely to be a combination of new, old, shared, borrowed or stolen code and adopted and adapted behaviours. Further, old malware can re-emerge after years in the wilderness, co-opted into an adversary’s latest arsenal to take defences by surprise.

Detection models need to ensure they can continue to detect older malware samples and not just the most recent ones.

Updating AI detection models

When it comes to updating AI detection models with new malware samples, vendors have a choice between two options.

The first is to keep a copy of every sample they might ever want to detect and retrain the model repeatedly on an ever-increasing volume of data. This results in better overall performance, but also slower updates and fewer releases.

The second is to only update the detection model on new samples. This is known as fine-tuning. During each step of the fine-tuning process, the model updates its understanding according to the new knowledge added and the impact of this on the patterns seen overall. As a result, the model can ‘forget’ the old patterns it learned previously (catastrophic forgetting). However, training a model on less data means the model updates faster and can be released more frequently, keeping better pace with the rapid rate of change of malware.

Regardless of the option chosen, the need to keep training AI detection models on new samples is critical.

The patterns that AI learns from malware samples enable it to generalise and detect not only what it was trained on, but also never before seen samples that bear at least some resemblance to the training data. Over time, however, new samples will begin to deviate enough that an old model’s effectiveness will decay and it will need to be updated.

The three detection update options evaluated by Hillary Sanders were:

1. Learning based on a selection of old and new samples

This is called ‘data-rehearsal’ and involves taking a small selection of old samples and mixing them in with the new, never-before-seen training data. Using this, the model is ‘reminded’ of the old information it needed to detect older samples, while at the same time learning to detect the newer ones.

2. Learning rate

This approach involves modifying how quickly the model learns by adjusting how much it can change after seeing any given sample. If the learning rate is too fast (in which case the model can change a lot with each sample added), it will only remember the most recent samples that it has seen. If the learning rate is too slow (the model can change only slightly with each sample added), it takes too long to learn anything. Finding the right trade-off between learning rate, retaining old information and adding new information can be tricky.

3. Elastic Weight Consolidation (EWC)

This approach was inspired by work by Google’s DeepMind in 2017 and it involves using the old model like an elastic spring to ‘pull back’ the new model if it starts to forget. For a more in-depth explanation of how to implement this approach, read Hillary Sanders’ blog post at https://ai.sophos.com/2021/02/02/catastrophic-forgetting-part-1/.

Findings

All three approaches performed better on older malware samples than on newer samples. Both the EWC and learning-rate approaches remove the need and cost of maintaining older data. However, the graph shows that while their future performance (using new data) is stronger than that achieved using the data-rehearsal technique, they don’t perform as well as data-rehearsal when it comes to remembering past data.

Because the data-rehearsal technique enables faster training and update releases, dips in future performance are more short term and therefore less worrying. Overall, the research showed that the data-rehearsal approach offers the best compromise between simplicity, update speed and performance in malware detection modelling.

Conclusion

In the malware detection game, being able to remember the past is almost as important as being able to predict the future. This must be balanced against the cost and speed of updating your model with new information. Data-rehearsal is a simple and effective way to protect the model’s ability to detect old malware while significantly increasing the pace at which you can update and release new models.

Read more at https://ai.sophos.com/




Share this article:
Share via emailShare via LinkedInPrint this page



Further reading:

Highest increase in global cyberattacks in two years
Information Security News & Events
Check Point Global Research released new data on Q2 2024 cyber-attack trends, noting a 30% global increase in Q2 2024, with Africa experiencing the highest average weekly per organisation.

Read more...
Upgrade your PCs to improve security
Information Security Infrastructure
Truly secure technology today must be designed to detect and address unusual activity as it happens, wherever it happens, right down to the BIOS and silicon levels.

Read more...
Open source code can also be open risk
Information Security Infrastructure
Software development has changed significantly over the years, and today, open-source code increasingly forms the foundation of modern applications, with surveys indicating that 60 – 90% of the average application's code base consists of open-source components.

Read more...
DeepSneak deception
Information Security News & Events
Kaspersky Global Research & Analysis researchers have discovered a new malicious campaign which is distributing a Trojan through a fake DeepSeek-R1 Large Language Model (LLM) app for PCs.

Read more...
SA’s strained, loadshedding-prone grid faces cyberthreats
Power Management Information Security
South Africa’s energy sector, already battered by decades of underinvestment and loadshedding, faces another escalating crisis; a wave of cyberthreats that could turn disruptions into catastrophic failures. Attacks are already happening internationally.

Read more...
Almost 50% of companies choose to pay the ransom
News & Events Information Security
This year’s Sophos State of Ransomware 2025 report found that nearly 50% of companies paid the ransom to get their data back, the second-highest rate of ransom payment for ransom demands in six years.

Read more...
Survey highlights cost of cyberdamage to industrial companies
Kaspersky Information Security News & Events
The majority of industrial organisations estimate their financial losses caused by cyberattacks to be over $1 million, while almost one in four report losses exceeding $5 million, and for some, it surpasses $10 million.

Read more...
Digital economy needs an agile approach to cybersecurity
Information Security News & Events
South Africa is the most targeted country in Africa when it comes to infostealer and ransomware attacks. Being at the forefront of the continent’s digital transformation puts South Africa in the crosshairs for sophisticated cyberattacks

Read more...
SIEM rule threat coverage validation
Information Security News & Events
New AI-detection engineering assistant from Cymulate automates SIEM rule validation for SecOps and blue teams by streamlining threat detection engineering with automated testing, control integrations and enhanced detections.

Read more...
Cybersecurity a challenge in digitalising OT
Kaspersky Information Security Industrial (Industry)
According to a study by Kaspersky and VDC Research on securing operational technology environments, the primary risks are inadequate security measures, insufficient resources allocated to OT cybersecurity, challenges surrounding regulatory compliance, and the complexities of IT/OT integration.

Read more...










While every effort has been made to ensure the accuracy of the information contained herein, the publisher and its agents cannot be held responsible for any errors contained, or any loss incurred as a result. Articles published do not necessarily reflect the views of the publishers. The editor reserves the right to alter or cut copy. Articles submitted are deemed to have been cleared for publication. Advertisements and company contact details are published as provided by the advertiser. Technews Publishing (Pty) Ltd cannot be held responsible for the accuracy or veracity of supplied material.




© Technews Publishing (Pty) Ltd. | All Rights Reserved.