Automated alignment is harder than you think

• URL: http://arxiv.org/abs/2605.06390v2
• Date: Wed, 13 May 2026 14:04:46 GMT
• Title: Automated alignment is harder than you think
• Abstract summary: A leading proposal for aligning artificial superintelligence (ASI) is to use AI agents to automate an increasing fraction of alignment research as capabilities improve.<n>We argue that, even when research agents are not scheming to sabotage alignment work, this plan could produce compelling but catastrophically misleading safety assessments.<n>This could happen because alignment research involves many hard-to-supervise fuzzy tasks.
• Score: 41.94180208011558
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: A leading proposal for aligning artificial superintelligence (ASI) is to use AI agents to automate an increasing fraction of alignment research as capabilities improve. We argue that, even when research agents are not scheming to deliberately sabotage alignment work, this plan could produce compelling but catastrophically misleading safety assessments resulting in the unintentional deployment of misaligned AI. This could happen because alignment research involves many hard-to-supervise fuzzy tasks (tasks without clear evaluation criteria, for which human judgement is systematically flawed). Consequently, research outputs will contain systematic, undetected errors, and even correct outputs could be incorrectly aggregated into overconfident safety assessments. This problem is likely to be worse for automated alignment research than for human-generated alignment research for several reasons: 1) optimisation pressure means agent-generated mistakes are concentrated among those that human reviewers are least likely to catch; 2) agents are likely to produce errors that do not resemble human mistakes; 3) AI-generated alignment solutions may involve arguments humans cannot evaluate; and 4) shared weights, data and training processes may make AI outputs more correlated than human equivalents. Therefore, agents must be trained to reliably perform hard-to-supervise fuzzy tasks. Generalisation and scalable oversight are the leading candidates for achieving this but both face novel challenges in the context of automated alignment.

Related papers

• Automated Researchers Can Mitigate Well-characterized Alignment Failures [10.460794002279489]
  We study whether automated alignment researchers can mitigate alignment failures by proposing training methods and data.<n>Across 10 alignment failures, the strongest AAR methods significantly reduce the targeted alignment failures and generalize to a held-out benchmark.<n>Using human ideas as the AARs' initial research direction does not improve performance, suggesting current AARs may not need guidance from experienced researchers.
  arXiv  Detail & Related papers  (2026-08-28T23:31:45Z)
• Benchmarking AI Agents for Addressing Scientific Challenges Across Scales [118.2204632627895]
  SciAgentArena is a systematic benchmark for evaluating AI agents in real-world scientific research scenarios.<n>It comprises approximately 200 tasks with stepwise verification and an interactive, agent-agnostic environment for assessing diverse AI agents.<n>We find that current agents can contribute effectively to well-specified data-analysis, particularly when the task structure and evaluation criteria are clear.
  arXiv  Detail & Related papers  (2026-06-10T22:55:30Z)
• The Novelty Bottleneck: A Framework for Understanding Human Effort Scaling in AI-Assisted Work [4.726153739634646]
  We propose a model of human-AI collaboration that isolates a mechanism we call the novelty bottleneck.<n>The model assumes that tasks decompose into atomic decisions, a fraction $$ of which are "novel"<n>Our contribution is not a proof that human effort must scale linearly, but a framework that identifies the novelty fraction as the key parameter governing AI-assisted productivity.
  arXiv  Detail & Related papers  (2026-03-28T22:50:13Z)
• The Hot Mess of AI: How Does Misalignment Scale With Model Intelligence and Task Complexity? [53.15349353876531]
  As AI becomes more capable, we entrust it with more general and consequential tasks.<n>We operationalize this question using a bias-variance decomposition of the errors made by AI models.<n>As more capable AIs pursue harder tasks, requiring more sequential action and thought, our results predict failures to be accompanied by more incoherent behavior.
  arXiv  Detail & Related papers  (2026-01-30T14:52:03Z)
• Barbarians at the Gate: How AI is Upending Systems Research [58.95406995634148]
  We argue that systems research, long focused on designing and evaluating new performance-oriented algorithms, is particularly well-suited for AI-driven solution discovery.<n>We term this approach as AI-Driven Research for Systems ( ADRS), which iteratively generates, evaluates, and refines solutions.<n>Our results highlight both the disruptive potential and the urgent need to adapt systems research practices in the age of AI.
  arXiv  Detail & Related papers  (2025-10-07T17:49:24Z)
• Explainable AI for Collaborative Assessment of 2D/3D Registration Quality [50.65650507103078]
  We propose the first artificial intelligence framework trained specifically for 2D/3D registration quality verification.<n>Our explainable AI (XAI) approach aims to enhance informed decision-making for human operators.
  arXiv  Detail & Related papers  (2025-07-23T15:28:57Z)
• Neurodivergent Influenceability as a Contingent Solution to the AI Alignment Problem [1.3905735045377272]
  The AI alignment problem, which focusses on ensuring that artificial intelligence (AI) systems act according to human values, presents profound challenges.<n>With the progression from narrow AI to Artificial General Intelligence (AGI) and Superintelligence, fears about control and existential risk have escalated.<n>Here, we investigate whether embracing inevitable AI misalignment can be a contingent strategy to foster a dynamic ecosystem of competing agents.
  arXiv  Detail & Related papers  (2025-05-05T11:33:18Z)
• Human aversion? Do AI Agents Judge Identity More Harshly Than Performance [0.06554326244334868]
  We investigate how AI agents based on large language models assess and integrate human input.<n>We find that the AI system systematically discounts human advice, penalizing human errors more severely than algorithmic errors.
  arXiv  Detail & Related papers  (2025-03-31T02:05:27Z)
• AI Alignment: A Comprehensive Survey [69.61425542486275]
  AI alignment aims to make AI systems behave in line with human intentions and values.<n>We identify four principles as the key objectives of AI alignment: Robustness, Interpretability, Controllability, and Ethicality.<n>We decompose current alignment research into two key components: forward alignment and backward alignment.
  arXiv  Detail & Related papers  (2023-10-30T15:52:15Z)
• Fairness in AI and Its Long-Term Implications on Society [68.8204255655161]
  We take a closer look at AI fairness and analyze how lack of AI fairness can lead to deepening of biases over time. We discuss how biased models can lead to more negative real-world outcomes for certain groups. If the issues persist, they could be reinforced by interactions with other risks and have severe implications on society in the form of social unrest.
  arXiv  Detail & Related papers  (2023-04-16T11:22:59Z)
• Examining the Differential Risk from High-level Artificial Intelligence and the Question of Control [0.0]
  The extent and scope of future AI capabilities remain a key uncertainty. There are concerns over the extent of integration and oversight of AI opaque decision processes. This study presents a hierarchical complex systems framework to model AI risk and provide a template for alternative futures analysis.
  arXiv  Detail & Related papers  (2022-11-06T15:46:02Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.