arXiv:2601.17355v1 Announce Type: new Abstract: Improvements in software defined networking allow for policy to be informed and modified by data-driven applications that can adjust policy to accommodate fluctuating requirements at line speed. However, there is some concern that over-correction can occur and cause unintended consequences depending on the data received. This is particularly problematic for network security features, such as machine-learning intrusion detection systems. We present Safeguard, a rule-based policy that overlaps a data-driven policy to prevent unintended responses for edge cases in network traffic. We develop a reference implementation of a network traffic classifier that enforces firewall rules for malicious traffic, and show how additional rulesets to allow kno…
arXiv:2601.17355v1 Announce Type: new Abstract: Improvements in software defined networking allow for policy to be informed and modified by data-driven applications that can adjust policy to accommodate fluctuating requirements at line speed. However, there is some concern that over-correction can occur and cause unintended consequences depending on the data received. This is particularly problematic for network security features, such as machine-learning intrusion detection systems. We present Safeguard, a rule-based policy that overlaps a data-driven policy to prevent unintended responses for edge cases in network traffic. We develop a reference implementation of a network traffic classifier that enforces firewall rules for malicious traffic, and show how additional rulesets to allow known-good traffic are essential in utilizing a data-driven network policy.