I was reading Cloudflare’s recent Introducing Adaptive Intelligence. the idea is a bot detection engine that continuously retrains on live traffic. it also proposes short-lived, disposable detection rules that appear and disappear at randomized intervals, making it harder for bot operators to learn which changes bypass a defense. continuous retraining was live at announcement time; disposable rules were still to follow.
A defense that keeps changing flips that calculation, but only if two things are true at once. First, it has to cost the defender less to react than it costs the attacker to work around it. Second, attackers must be starved of the feedback they use to adapt, so they cannot simply learn their way back in.
Chris Pope, cloudflare: Introducing Adaptive Intelligence.
i think the triage time for the attacker is the most interesting cost here. when a previously working attack stops working, a human may have to figure out why. that attention is expensive when scaling. an unfamiliar codebase, more agent runs, and reviewing execution traces can all add to the cost. the agent might resolve it independently, but in my own workflow i tend to step in initially.
this fits alongside token burn. a defender may be able to introduce a relatively cheap change that takes considerably more effort to diagnose. it does not have to stop every attack or force human intervention every time to be useful.
cloudflare’s proposed disposable rules could make that diagnosis harder. if a rule disappears while an agent is adjusting its attack, the next attempt might succeed without the adjustment having fixed anything. the operator could walk away with the wrong explanation for what worked.
there are no measured triage costs in the post. but making an attacker revisit something they thought was working is a practical direction for active defense, especially as inference gets cheaper.