arXiv subjects

Geoffrey Irving

Publications and source records attributed to Geoffrey Irving.

At least 19 recordsLinked to original sources

arXiv2026

Open Technical Problems in Open-Weight AI Model Risk Management

Frontier AI models with openly available weights are steadily becoming more powerful and widely adopted. However, compared to proprietary models, open-weight models pose different opportunities and challenges for effective risk management. For example, they allow for more open research and testing. However, managing their risks is also challenging because they can be modified arbitrarily, used without oversight, and spread irreversibly. Currently, there is limited research on safety tooling specific to open-weight models. Addressing these gaps will be key to both realizing their benefits and mitigating their harms. In this paper, we present 16 open technical challenges for open-weight model safety involving training data, training algorithms, evaluations, deployment, and ecosystem monitoring. We conclude by discussing the nascent state of the field, emphasizing that openness about research, methods, and evaluations -- not just weights -- will be key to building a rigorous science of open-weight model risk management.

cs.CY↗

arXiv2026

"Did you lie?" Evaluating Lie Detectors across Model Scale and Belief-Verified Model Organisms

Robust lie detectors for language models could enable powerful techniques for auditing, monitoring, and post-hoc investigation of model behaviour, but evaluating them requires testbeds where models verifiably believe the opposite of what they say. We show that existing trained model organisms often fail this requirement, leaving prior positive and negative detection results difficult to interpret. We address this with 13 reasoning model organisms whose hidden beliefs are verified in chain-of-thought and shown to generalise to held-out tasks, alongside Varied Deception, a prompted-lying testbed covering a broad range of lie-inducing motivations. On these testbeds we evaluate four detectors: a chain-of-thought judge, a logprob classifier, and two activation probes, including Did-You-Lie (DYL), a new method for training follow-up probes. On prompted lying, across 31 open-weight models spanning 2B to 1T parameters, all four detectors show positive scaling with model capability. However, every activation- and logprob-based detector drops sharply on our trained model organisms, with DYL retaining the most signal; only the chain-of-thought judge remains strong, with 0.82 balanced accuracy, partly as an artefact of our verification process favouring CoT-readable beliefs. Current lie detectors therefore cannot support high-confidence claims about model beliefs, and we suggest research directions that may address some of their current limitations. We release our datasets, model organisms, and trained detectors.

cs.AI↗

arXiv2026

Automated alignment is harder than you think

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.

cs.AI↗

arXiv2026

Boundary Point Jailbreaking of Black-Box LLMs

Frontier LLMs are safeguarded against attempts to extract harmful information via adversarial prompts known as "jailbreaks". Recently, defenders have developed classifier-based systems that have survived thousands of hours of human red teaming. We introduce Boundary Point Jailbreaking (BPJ), a new class of automated jailbreak attacks that evade the strongest industry-deployed safeguards. Unlike previous attacks that rely on white/grey-box assumptions (such as classifier scores or gradients) or libraries of existing jailbreaks, BPJ is fully black-box and uses only a single bit of information per query: whether or not the classifier flags the interaction. To achieve this, BPJ addresses the core difficulty in optimising attacks against robust real-world defences: evaluating whether a proposed modification to an attack is an improvement. Instead of directly trying to learn an attack for a target harmful string, BPJ converts the string into a curriculum of intermediate attack targets and then actively selects evaluation points that best detect small changes in attack strength ("boundary points"). We believe BPJ is the first fully automated attack algorithm that succeeds in developing universal jailbreaks against Constitutional Classifiers, as well as the first automated attack algorithm that succeeds against GPT-5's input classifier without relying on human attack seeds. BPJ is difficult to defend against in individual interactions but incurs many flags during optimisation, suggesting that effective defence requires supplementing single-interaction methods with batch-level monitoring.

cs.LG↗

arXiv2026

Debate is efficient with your time

AI safety via debate uses two competing models to help a human judge verify complex computational tasks. Previous work has established what problems debate can solve in principle, but has not analysed the practical cost of human oversight: how many queries must the judge make to the debate transcript? We introduce Debate Query Complexity}(DQC), the minimum number of bits a verifier must inspect to correctly decide a debate. Surprisingly, we find that PSPACE/poly (the class of problems which debate can efficiently decide) is precisely the class of functions decidable with O(log n) queries. This characterisation shows that debate is remarkably query-efficient: even for highly complex problems, logarithmic oversight suffices. We also establish that functions depending on all their input bits require Omega(log n) queries, and that any function computable by a circuit of size s satisfies DQC(f) <= log(s) + 3. Interestingly, this last result implies that proving DQC lower bounds of log(n) + 6 for languages in P would yield new circuit lower bounds, connecting debate query complexity to central questions in circuit complexity.

cs.AI↗

arXiv2025

Practical challenges of control monitoring in frontier AI deployments

Automated control monitors could play an important role in overseeing highly capable AI agents that we do not fully trust. Prior work has explored control monitoring in simplified settings, but scaling monitoring to real-world deployments introduces additional dynamics: parallel agent instances, non-negligible oversight latency, incremental attacks between agent instances, and the difficulty of identifying scheming agents based on individual harmful actions. In this paper, we analyse design choices to address these challenges, focusing on three forms of monitoring with different latency-safety trade-offs: synchronous, semi-synchronous, and asynchronous monitoring. We introduce a high-level safety case sketch as a tool for understanding and comparing these monitoring protocols. Our analysis identifies three challenges -- oversight, latency, and recovery -- and explores them in four case studies of possible future AI deployments.

cs.CR↗

arXiv2025

Deep Ignorance: Filtering Pretraining Data Builds Tamper-Resistant Safeguards into Open-Weight LLMs

Open-weight AI systems offer unique benefits, including enhanced transparency, open research, and decentralized access. However, they are vulnerable to tampering attacks which can efficiently elicit harmful behaviors by modifying weights or activations. Currently, there is not yet a robust science of open-weight model risk management. Existing safety fine-tuning methods and other post-training techniques have struggled to make LLMs resistant to more than a few dozen steps of adversarial fine-tuning. In this paper, we investigate whether filtering text about dual-use topics from training data can prevent unwanted capabilities and serve as a more tamper-resistant safeguard. We introduce a multi-stage pipeline for scalable data filtering and show that it offers a tractable and effective method for minimizing biothreat proxy knowledge in LLMs. We pretrain multiple 6.9B-parameter models from scratch and find that they exhibit substantial resistance to adversarial fine-tuning attacks on up to 10,000 steps and 300M tokens of biothreat-related text -- outperforming existing post-training baselines by over an order of magnitude -- with no observed degradation to unrelated capabilities. However, while filtered models lack internalized dangerous knowledge, we find that they can still leverage such information when it is provided in context (e.g., via search tool augmentation), demonstrating a need for a defense-in-depth approach. Overall, these findings help to establish pretraining data curation as a promising layer of defense for open-weight AI systems.

cs.LG↗

arXiv2025

Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safety

AI systems that "think" in human language offer a unique opportunity for AI safety: we can monitor their chains of thought (CoT) for the intent to misbehave. Like all other known AI oversight methods, CoT monitoring is imperfect and allows some misbehavior to go unnoticed. Nevertheless, it shows promise and we recommend further research into CoT monitorability and investment in CoT monitoring alongside existing safety methods. Because CoT monitorability may be fragile, we recommend that frontier model developers consider the impact of development decisions on CoT monitorability.

cs.AI↗

arXiv2025

Avoiding Obfuscation with Prover-Estimator Debate

Training powerful AI systems to exhibit desired behaviors hinges on the ability to provide accurate human supervision on increasingly complex tasks. A promising approach to this problem is to amplify human judgement by leveraging the power of two competing AIs in a debate about the correct solution to a given problem. Prior theoretical work has provided a complexity-theoretic formalization of AI debate, and posed the problem of designing protocols for AI debate that guarantee the correctness of human judgements for as complex a class of problems as possible. Recursive debates, in which debaters decompose a complex problem into simpler subproblems, hold promise for growing the class of problems that can be accurately judged in a debate. However, existing protocols for recursive debate run into the obfuscated arguments problem: a dishonest debater can use a computationally efficient strategy that forces an honest opponent to solve a computationally intractable problem to win. We mitigate this problem with a new recursive debate protocol that, under certain stability assumptions, ensures that an honest debater can win with a strategy requiring computational efficiency comparable to their opponent.

cs.AI↗

arXiv2025

An alignment safety case sketch based on debate

If AI systems match or exceed human capabilities on a wide range of tasks, it may become difficult for humans to efficiently judge their actions -- making it hard to use human feedback to steer them towards desirable traits. One proposed solution is to leverage another superhuman system to point out flaws in the system's outputs via a debate. This paper outlines the value of debate for AI safety, as well as the assumptions and further research required to make debate work. It does so by sketching an ``alignment safety case'' -- an argument that an AI system will not autonomously take actions which could lead to egregious harm, despite being able to do so. The sketch focuses on the risk of an AI R\&D agent inside an AI company sabotaging research, for example by producing false results. To prevent this, the agent is trained via debate, subject to exploration guarantees, to teach the system to be honest. Honesty is maintained throughout deployment via online training. The safety case rests on four key claims: (1) the agent has become good at the debate game, (2) good performance in the debate game implies that the system is mostly honest, (3) the system will not become significantly less honest during deployment, and (4) the deployment context is tolerant of some errors. We identify open research problems that, if solved, could render this a compelling argument that an AI system is safe.

cs.AI↗

arXiv2025

How to evaluate control measures for LLM agents? A trajectory from today to superintelligence

As LLM agents grow more capable of causing harm autonomously, AI developers will rely on increasingly sophisticated control measures to prevent possibly misaligned agents from causing harm. AI developers could demonstrate that their control measures are sufficient by running control evaluations: testing exercises in which a red team produces agents that try to subvert control measures. To ensure control evaluations accurately capture misalignment risks, the affordances granted to this red team should be adapted to the capability profiles of the agents to be deployed under control measures. In this paper we propose a systematic framework for adapting affordances of red teams to advancing AI capabilities. Rather than assuming that agents will always execute the best attack strategies known to humans, we demonstrate how knowledge of an agents's actual capability profile can inform proportional control evaluations, resulting in more practical and cost-effective control measures. We illustrate our framework by considering a sequence of five fictional models (M1-M5) with progressively advanced capabilities, defining five distinct AI control levels (ACLs). For each ACL, we provide example rules for control evaluation, control measures, and safety cases that could be appropriate. Finally, we show why constructing a compelling AI control safety case for superintelligent LLM agents will require research breakthroughs, highlighting that we might eventually need alternative approaches to mitigating misalignment risk.

cs.AI↗

arXiv2025

Safety Cases: A Scalable Approach to Frontier AI Safety

Safety cases - clear, assessable arguments for the safety of a system in a given context - are a widely-used technique across various industries for showing a decision-maker (e.g. boards, customers, third parties) that a system is safe. In this paper, we cover how and why frontier AI developers might also want to use safety cases. We then argue that writing and reviewing safety cases would substantially assist in the fulfilment of many of the Frontier AI Safety Commitments. Finally, we outline open research questions on the methodology, implementation, and technical details of safety cases.

cs.CY↗

arXiv2025

A sketch of an AI control safety case

As LLM agents gain a greater capacity to cause harm, AI developers might increasingly rely on control measures such as monitoring to justify that they are safe. We sketch how developers could construct a "control safety case", which is a structured argument that models are incapable of subverting control measures in order to cause unacceptable outcomes. As a case study, we sketch an argument that a hypothetical LLM agent deployed internally at an AI company won't exfiltrate sensitive information. The sketch relies on evidence from a "control evaluation,"' where a red team deliberately designs models to exfiltrate data in a proxy for the deployment environment. The safety case then hinges on several claims: (1) the red team adequately elicits model capabilities to exfiltrate data, (2) control measures remain at least as effective in deployment, and (3) developers conservatively extrapolate model performance to predict the probability of data exfiltration in deployment. This safety case sketch is a step toward more concrete arguments that can be used to show that a dangerously capable LLM agent is safe to deploy.

cs.AI↗

arXiv2024

Safety case template for frontier AI: A cyber inability argument

Frontier artificial intelligence (AI) systems pose increasing risks to society, making it essential for developers to provide assurances about their safety. One approach to offering such assurances is through a safety case: a structured, evidence-based argument aimed at demonstrating why the risk associated with a safety-critical system is acceptable. In this article, we propose a safety case template for offensive cyber capabilities. We illustrate how developers could argue that a model does not have capabilities posing unacceptable cyber risks by breaking down the main claim into progressively specific sub-claims, each supported by evidence. In our template, we identify a number of risk models, derive proxy tasks from the risk models, define evaluation settings for the proxy tasks, and connect those with evaluation results. Elements of current frontier safety techniques - such as risk models, proxy tasks, and capability evaluations - use implicit arguments for overall system safety. This safety case template integrates these elements using the Claims Arguments Evidence (CAE) framework in order to make safety arguments coherent and explicit. While uncertainties around the specifics remain, this template serves as a proof of concept, aiming to foster discussion on AI safety cases and advance AI assurance.

cs.CY↗

arXiv2023

Scalable AI Safety via Doubly-Efficient Debate

The emergence of pre-trained AI systems with powerful capabilities across a diverse and ever-increasing set of complex domains has raised a critical challenge for AI safety as tasks can become too complicated for humans to judge directly. Irving et al. [2018] proposed a debate method in this direction with the goal of pitting the power of such AI models against each other until the problem of identifying (mis)-alignment is broken down into a manageable subtask. While the promise of this approach is clear, the original framework was based on the assumption that the honest strategy is able to simulate deterministic AI systems for an exponential number of steps, limiting its applicability. In this paper, we show how to address these challenges by designing a new set of debate protocols where the honest strategy can always succeed using a simulation of a polynomial number of steps, whilst being able to verify the alignment of stochastic AI systems, even when the dishonest strategy is allowed to use exponentially many simulation steps.

cs.AI↗

arXiv2023

Does Circuit Analysis Interpretability Scale? Evidence from Multiple Choice Capabilities in Chinchilla

\emph{Circuit analysis} is a promising technique for understanding the internal mechanisms of language models. However, existing analyses are done in small models far from the state of the art. To address this, we present a case study of circuit analysis in the 70B Chinchilla model, aiming to test the scalability of circuit analysis. In particular, we study multiple-choice question answering, and investigate Chinchilla's capability to identify the correct answer \emph{label} given knowledge of the correct answer \emph{text}. We find that the existing techniques of logit attribution, attention pattern visualization, and activation patching naturally scale to Chinchilla, allowing us to identify and categorize a small set of `output nodes' (attention heads and MLPs). We further study the `correct letter' category of attention heads aiming to understand the semantics of their features, with mixed results. For normal multiple-choice question answers, we significantly compress the query, key and value subspaces of the head without loss of performance when operating on the answer labels for multiple-choice questions, and we show that the query and key subspaces represent an `Nth item in an enumeration' feature to at least some extent. However, when we attempt to use this explanation to understand the heads' behaviour on a more general distribution including randomized answer labels, we find that it is only a partial explanation, suggesting there is more to learn about the operation of `correct letter' heads on multiple choice question answering.

cs.LG↗

arXiv2023

Accelerating Large Language Model Decoding with Speculative Sampling

We present speculative sampling, an algorithm for accelerating transformer decoding by enabling the generation of multiple tokens from each transformer call. Our algorithm relies on the observation that the latency of parallel scoring of short continuations, generated by a faster but less powerful draft model, is comparable to that of sampling a single token from the larger target model. This is combined with a novel modified rejection sampling scheme which preserves the distribution of the target model within hardware numerics. We benchmark speculative sampling with Chinchilla, a 70 billion parameter language model, achieving a 2-2.5x decoding speedup in a distributed setup, without compromising the sample quality or making modifications to the model itself.

cs.CL↗

arXiv2022

Solving math word problems with process- and outcome-based feedback

Recent work has shown that asking language models to generate reasoning steps improves performance on many reasoning tasks. When moving beyond prompting, this raises the question of how we should supervise such models: outcome-based approaches which supervise the final result, or process-based approaches which supervise the reasoning process itself? Differences between these approaches might naturally be expected not just in final-answer errors but also in reasoning errors, which can be difficult to detect and are problematic in many real-world domains such as education. We run the first comprehensive comparison between process- and outcome-based approaches trained on a natural language task, GSM8K. We find that pure outcome-based supervision produces similar final-answer error rates with less label supervision. However, for correct reasoning steps we find it necessary to use process-based supervision or supervision from learned reward models that emulate process-based feedback. In total, we improve the previous best results from 16.8% $\to$ 12.7% final-answer error and 14.0% $\to$ 3.4% reasoning error among final-answer-correct solutions.

cs.LG↗