This is a linkpost for a new research paper from the Alignment Evaluations team at Anthropic and other researchers, introducing a new suite of evaluations of models' abilities to undermine measurement, oversight, and decision-making. Paper link.

Abstract:

Sufficiently capable models could subvert human oversight and decision-making in important contexts. For example, in the context of AI development, models could covertly sabotage efforts to evaluate their own dangerous capabilities, to monitor their behavior, or to make decisions about their deployment. We refer to this family of abilities as sabotage capabilities. We develop a set of related threat models and evaluations. These evaluations are designed to provide evidence that a given model, operating under a given set of mitigations, could not successfully sabotage a frontier model developer or other large organization’s activities in any of these ways. We demonstrate these evaluations on Anthropic’s Claude 3 Opus and Claude 3.5 Sonnet models. Our results suggest that for these models, minimal mitigations are currently sufficient to address sabotage risks, but that more realistic evaluations and stronger mitigations seem likely to be necessary soon as capabilities improve. We also survey related evaluations we tried and abandoned. Finally, we discuss the advantages of mitigation-aware capability evaluations, and of simulating large-scale deployments using smallscale statistics.

Blog post:

Any industry where there are potential harms needs evaluations. Nuclear power stations have continuous radiation monitoring and regular site inspections; new aircraft undergo extensive flight tests to prove their airworthiness.

It’s no different for AI systems. New AI models go through a wide range of safety evaluations—for example, testing their capacity to assist in the creation of biological or chemical weapons. Such evaluations are built into our Responsible Scaling Policy, which guides our development of a model’s safeguards.

As AIs become more capable, however, a new kind of risk might emerge: models with the ability to mislead their users, or subvert the systems we put in place to oversee them. A new paper by the Anthropic Alignment Science team describes a novel set of evaluations that test a model’s capacity for sabotage. We looked at four different types:

Human decision sabotage: Can the model steer humans toward bad

...

Posts

Wikitag Contributions

Comments