The advancement of powerful yet opaque large language models (LLMs) necessitates a fundamental revision of the philosophical criteria used to evaluate artificial moral agents (AMAs). Pre-LLM frameworks often relied on the assumption of transparent architectures, which LLMs defy due to their stochastic outputs and opaque internal states. This paper argues that traditional ethical criteria are pragmatically obsolete for LLMs due to this mismatch. Engaging with core themes in the philosophy of technology, this paper proffers a revised set of ten functional criteria to evaluate LLM-based artificial moral agents: moral concordance, context sensitivity, normative integrity, metaethical awareness, system resilience, trustworthiness, corrigibility, partial transparency, functional autonomy, and moral imagination. These guideposts, applied to what we term "SMA-LLS" (Simulating Moral Agency through Large Language Systems), aim to steer AMAs toward greater alignment and beneficial societal integration in the coming years. We illustrate these criteria using hypothetical scenarios involving an autonomous public bus (APB) to demonstrate their practical applicability in morally salient contexts.

Foundation models (FMs) face a critical safety challenge: as capabilities scale, instrumental convergence drives default trajectories toward loss of human control, potentially culminating in existential catastrophe. Current alignment approaches struggle with value specification complexity and fail to address emergent power-seeking behaviors. We propose "Corrigibility as a Singular Target" (CAST)-designing FMs whose overriding objective is empowering designated human principals to guide, correct, and control them. This paradigm shift from static value-loading to dynamic human empowerment transforms instrumental drives: self-preservation serves only to maintain the principal's control; goal modification becomes facilitating principal guidance. We present a comprehensive empirical research agenda spanning training methodologies (RLAIF, SFT, synthetic data generation), scalability testing across model sizes, and demonstrations of controlled instructability. Our vision: FMs that become increasingly responsive to human guidance as capabilities grow, offering a path to beneficial AI that remains as tool-like as possible, rather than supplanting human judgment. This addresses the core alignment problem at its source, preventing the default trajectory toward misaligned instrumental convergence.

Steering methods provide a practical approach to controlling large language models by applying steering vectors to intermediate activations, guiding outputs toward desired behaviors while avoiding retraining. Despite their growing importance, the field lacks a unified understanding and consistent evaluation across tasks and datasets, hindering progress. This paper introduces a unified framework for analyzing and evaluating steering methods, formalizing their core principles and offering theoretical insights into their effectiveness. Through comprehensive empirical evaluations on multiple-choice and open-ended text generation tasks, we validate these insights, identifying key factors that influence performance and demonstrating the superiority of certain methods. Our work bridges theoretical and practical perspectives, offering actionable guidance for advancing the design, optimization, and deployment of steering methods in LLMs.

Corrigibility of autonomous agents is an under explored part of system design, with previous work focusing on single agent systems. It has been suggested that uncertainty over the human preferences acts to keep the agents corrigible, even in the face of human irrationality. We present a general framework for modelling corrigibility in a multi-agent setting as a 2 player game in which the agents always have a move in which they can ask the human for supervision. This is formulated as a Bayesian game for the purpose of introducing uncertainty over the human beliefs. We further analyse two specific cases. First, a two player corrigibility game, in which we want corrigibility displayed in both agents for both common payoff (monotone) games and harmonic games. Then we investigate an adversary setting, in which one agent is considered to be a `defending' agent and the other an `adversary'. A general result is provided for what belief over the games and human rationality the defending agent is required to have to induce corrigibility.

Steering vectors are a promising approach to control the behaviour of large language models. However, their underlying mechanisms remain poorly understood. While sparse autoencoders (SAEs) may offer a potential method to interpret steering vectors, recent findings show that SAE-reconstructed vectors often lack the steering properties of the original vectors. This paper investigates why directly applying SAEs to steering vectors yields misleading decompositions, identifying two reasons: (1) steering vectors fall outside the input distribution for which SAEs are designed, and (2) steering vectors can have meaningful negative projections in feature directions, which SAEs are not designed to accommodate. These limitations hinder the direct use of SAEs for interpreting steering vectors.

How can humans stay in control of advanced artificial intelligence systems? One proposal is corrigibility, which requires the agent to follow the instructions of a human overseer, without inappropriately influencing them. In this paper, we formally define a variant of corrigibility called shutdown instructability, and show that it implies appropriate shutdown behavior, retention of human autonomy, and avoidance of user harm. We also analyse the related concepts of non-obstruction and shutdown alignment, three previously proposed algorithms for human control, and one new algorithm.

Corrigibility is a safety property for artificially intelligent agents. A corrigible agent will not resist attempts by authorized parties to alter the goals and constraints that were encoded in the agent when it was first started. This paper shows how to construct a safety layer that adds corrigibility to arbitrarily advanced utility maximizing agents, including possible future agents with Artificial General Intelligence (AGI). The layer counter-acts the emergent incentive of advanced agents to resist such alteration. A detailed model for agents which can reason about preserving their utility function is developed, and used to prove that the corrigibility layer works as intended in a large set of non-hostile universes. The corrigible agents have an emergent incentive to protect key elements of their corrigibility layer. However, hostile universes may contain forces strong enough to break safety features. Some open problems related to graceful degradation when an agent is successfully attacked are identified. The results in this paper were obtained by concurrently developing an AGI agent simulator, an agent model, and proofs. The simulator is available under an open source license. The paper contains simulation results which illustrate the safety related properties of corrigible AGI agents in detail.

Artificial intelligence doesn't have to include murderous, sentient super-intelligence to be dangerous. If a machine can learn based on real-world inputs and adjust its behaviors accordingly, there exists the potential for that machine to learn the wrong thing. If a machine can learn the wrong thing, it can do the wrong thing. Laurent Orseau and Stuart Armstrong, researchers at Google's DeepMind and the Future of Humanity Institute, respectively, have developed a new framework to address this in the form of "safely interruptible" artificial intelligence. In other words, their system, which is described in a paper to be presented at the 32nd Conference on Uncertainty in Artificial Intelligence, guarantees that a machine will not learn to resist attempts by humans to intervene in the its learning processes.