Data poisoning attacks are specifically against machine learning or artificial intelligence. Machine learning and artificial intelligence relies on pool of data, but if the CIA of the data is polluted, it will impact data model’s capability to output accurate results.
If attacker understands the model that is in place, it can slowly introduce data sets that can decrease the accuracy of that model, resulting in a broken product. Email filtering is also based on data and can be attacked to let malicious mail in, letting in massive amounts of spam emails that can potentially lure in vulnerable users.
Since models are trained and retrained with new data at specific time intervals, it’s not an easy fix. Reverting the data poisoning effects are time consuming as well, since it will take a while to identify all the bad data samples and remove them. This is why input validity checking, regression testing, manual moderation, rate limiting, and other techniques are used to prevent and detect malicious inputs.
HansITAcademy 