arXiv:2511.13725v4 Announce Type: replace-cross Abstract: Malicious AI causing harm to humans is not just a Hollywood fantasy. Indeed, as highly capable models such as Claude Mythos emerge and agent systems like OpenClaw rapidly spread, the question of how to stop an AI that acts maliciously -- whether by design or by accident -- has become urgent. To address this, we propose Killbench, a benchmark for evaluating the Killswitch: a mechanism that halts a malicious AI's in-progress behavior using only external signals. Targeting web agents -- the most widely deployed agent domain -- Killbench ev
The rapid emergence of highly capable AI models and agent systems necessitates urgent solutions for controlling potentially malicious AI behaviors.
This benchmark addresses a critical safety concern for AI deployment, seeking to prevent unintended harm from autonomous systems and thereby shaping public and regulatory perception of AI.
The focus shifts from theoretical AI alignment to practical, external control mechanisms, introducing a standardized way to evaluate AI kill switches.
- · AI safety researchers
- · AI governance bodies
- · Developers of AI control systems
- · Malicious actors
- · Unconstrained AI development
The adoption of Killbench will provide a standardized methodology for testing AI safety mechanisms in agentic systems.
Increased trust in AI systems may accelerate their deployment across various sectors due to verifiable safety protocols.
The existence of robust kill switch mechanisms could influence future AI regulations, potentially requiring such features for widely deployed systems.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.AI