5 Artificial Intelligence and Machine Learning Use-Cases for Cybersecurity
Guest postsGuest post2 min read
The five areas where artificial intelligence and machine learning already help security teams are listed below, each paired with a documented deployment or study. The common thread is scale: software sifts billions of log lines, emails and network events that no human team could review, then hands analysts a short list of genuine threats.
| Use-case | How AI/ML is applied | Real-world reference | Reported figure |
|---|---|---|---|
| 1. Risk identification | Models trained on past incidents point to misconfigurations, weak passwords and patching gaps before attackers find them | IBM Security Services 2014 Cyber Security Intelligence Index, covering about 1,000 clients in 133 countries | Human error contributed to more than 95% of the incidents IBM investigated |
| 2. Early detection | Unsupervised learning builds a baseline of normal behaviour for each user and device, then flags deviations | Darktrace Enterprise Immune System, from a company founded in Cambridge, UK, in 2013 | Needs no prior signatures; Gartner noted it lacks a known-threat detection capability |
| 3. Round-the-clock operational screening | Classifiers check every incoming message and link, warning users at the moment they click | Gmail filters linked to Google Safe Browsing, announced 31 May 2017 | 99.9% accuracy on spam and phishing; spam makes up 50 to 70% of Gmail traffic |
| 4. Situational awareness | The system ranks the most abnormal events so analysts label only the riskiest, and the model learns from their answers | AI² by MIT CSAIL and PatternEx, presented in April 2016 | Caught 85% of attacks, about three times earlier benchmarks, with five times fewer false positives |
| 5. Corrective action | Automated tools shorten detection, remediation and patching after an alert | Capgemini Research Institute survey of 850 senior executives in 10 countries, July 2019 | 74% said AI speeds up response, trimming those times by 12% on average |
The numbers behind the case
| Indicator | Figure | Source and date |
|---|---|---|
| Data used to test AI² | 3.6 billion log lines from millions of users over three months | MIT News, April 2016 |
| Analyst workload with AI² | 200 events to review on day one, dropping to 30 to 40 a day | MIT News, April 2016 |
| Suspicious Gmail messages held for deeper checks | About 0.05% of mail | TechCrunch, May 2017 |
| Attacks on IBM client firewalls | Nearly 2 million a week; 17,000 serious attempts found among 91 million events | Security Magazine, June 2014 |
| Executives who say they cannot respond to attacks without AI | 69% | Capgemini, July 2019 |
| Executives who say AI makes breach detection more accurate | 69% | Capgemini, July 2019 |
| Firms whose analysts are overwhelmed by data volume | 56% | Capgemini, July 2019 |
| AI adoption in security | One in five firms before 2019; 63% planned it by 2020 | Capgemini, July 2019 |
| Forecast global cost of data breaches by 2019 | $2.1 trillion | Juniper Research, reported by IT Pro, May 2015 |
Sources: MIT News, TechCrunch, eSecurity Planet, Security Magazine, Capgemini, IT Pro.
This column was written by an outside contributor. Views are the author's own and do not necessarily reflect those of Bit News Today.

