Tobias Losch

What you will learn:

1. AI scales the fight, but it does not flatten the field. Both attackers and defenders gain from AI — but the gains are not symmetric. Adversaries use AI to reduce the cost of launching abuse at volume. Defenders use AI to reduce the cost of processing it. What this produces is not an arms race that reaches equilibrium; it is a higher-volume contest in which the structural advantages of the offense — speed, anonymity, disposable infrastructure — remain intact. AI accelerates throughput on both sides without closing the gap. 2. The friction points in remediation are not technical — and AI won't dissolve them. The bottlenecks in abuse remediation sit in evidence qualification, provider decision-making, and jurisdictional enforcement — none of which AI controls. A better-drafted takedown submission still waits on a provider's review queue. A more precisely scoped abuse claim still requires a human determination by a registrar or platform. The regulatory frameworks that govern what constitutes actionable abuse are not adapting at the same rate as the threats they were designed to address. AI improves the inputs; it does not change the gatekeeping structure. 3. Uneven adoption creates new asymmetries — even among defenders. AI is not being implemented uniformly across security vendors, providers, or enforcement bodies. The speed gap between a defender with mature AI-assisted workflows and one without is growing. The same is true for providers: some are investing in AI-assisted abuse review, others are not. This fragmentation means the ecosystem cannot be evaluated as a single system moving in one direction. Progress in one part does not translate to progress across the chain — and defenders who assume otherwise will build strategies that overestimate their actual reach.