English
Related papers

Related papers: Sabotage Evaluations for Frontier Models

200 papers

We evaluate the propensity of frontier models to sabotage or refuse to assist with safety research when deployed as AI research agents within a frontier AI company. We apply two complementary evaluations to four Claude models (Mythos…

Artificial Intelligence · Computer Science 2026-04-28 Robert Kirk , Alexandra Souly , Kai Fronsdal , Abby D'Cruz , Xander Davies

This technical report presents methods developed by the UK AI Security Institute for assessing whether advanced AI systems reliably follow intended goals. Specifically, we evaluate whether frontier models sabotage safety research when…

Artificial Intelligence · Computer Science 2026-04-02 Alexandra Souly , Robert Kirk , Jacob Merizian , Abby D'Cruz , Xander Davies

Trustworthy capability evaluations are crucial for ensuring the safety of AI systems, and are becoming a key component of AI regulation. However, the developers of an AI system, or the AI system itself, may have incentives for evaluations…

Artificial Intelligence · Computer Science 2025-02-10 Teun van der Weij , Felix Hofstätter , Ollie Jaffe , Samuel F. Brown , Francis Rhys Ward

Recent work has demonstrated the plausibility of frontier AI models scheming -- knowingly and covertly pursuing an objective misaligned with its developer's intentions. Such behavior could be very hard to detect, and if present in future…

Machine Learning · Computer Science 2025-07-04 Mary Phuong , Roland S. Zimmermann , Ziyue Wang , David Lindner , Victoria Krakovna , Sarah Cogan , Allan Dafoe , Lewis Ho , Rohin Shah

To understand the risks posed by a new AI system, we must understand what it can and cannot do. Building on prior work, we introduce a programme of new "dangerous capability" evaluations and pilot them on Gemini 1.0 models. Our evaluations…

Frontier models are increasingly trained and deployed as autonomous agent. One safety concern is that AI agents might covertly pursue misaligned goals, hiding their true capabilities and objectives - also known as scheming. We study whether…

Artificial Intelligence · Computer Science 2025-01-16 Alexander Meinke , Bronson Schoen , Jérémy Scheurer , Mikita Balesni , Rusheb Shah , Marius Hobbhahn

Pre-deployment evaluations inspect only a limited sample of model actions. A malicious model seeking to evade oversight could exploit this by randomizing when to "defect": misbehaving so rarely that no malicious actions are observed during…

Machine Learning · Computer Science 2026-03-03 Alex Serrano , Wen Xing , David Lindner , Erik Jenner

AI systems are increasingly able to autonomously conduct realistic software engineering tasks, and may soon be deployed to automate machine learning (ML) R&D itself. Frontier AI systems may be deployed in safety-critical settings, including…

Current approaches to building general-purpose AI systems tend to produce systems with both beneficial and harmful capabilities. Further progress in AI development could lead to capabilities that pose extreme risks, such as offensive cyber…

Today's leading AI models engage in sophisticated behaviour when placed in strategic competition. They spontaneously attempt deception, signaling intentions they do not intend to follow; they demonstrate rich theory of mind, reasoning about…

Artificial Intelligence · Computer Science 2026-02-17 Kenneth Payne

Developers try to evaluate whether an AI system can be misused by adversaries before releasing it; for example, they might test whether a model enables cyberoffense, user manipulation, or bioterrorism. In this work, we show that…

Cryptography and Security · Computer Science 2024-07-03 Erik Jones , Anca Dragan , Jacob Steinhardt

AI leaders and safety reports increasingly warn that advances in model reasoning may enable biological misuse, including by low-expertise users, while major labs describe safeguards as expanding but still evolving rather than settled. This…

Computers and Society · Computer Science 2026-04-24 Michael Richter

AI evaluations are an important component of the AI governance toolkit, underlying current approaches to safety cases for preventing catastrophic risks. Our paper examines what these evaluations can and cannot tell us. Evaluations can…

Computers and Society · Computer Science 2024-12-13 Peter Barnett , Lisa Thiergart

Trustworthy evaluations of dangerous capabilities are increasingly crucial for determining whether an AI system is safe to deploy. One empirically demonstrated threat is sandbagging - the strategic underperformance on evaluations by AI…

Cryptography and Security · Computer Science 2025-11-03 Chloe Li , Mary Phuong , Noah Y. Siegel

Evaluating the safety of frontier AI systems is an increasingly important concern, helping to measure the capabilities of such models and identify risks before deployment. However, it has been recognised that if AI agents are aware that…

Machine Learning · Computer Science 2025-10-01 Joel Dyer , Daniel Jarne Ornia , Nicholas Bishop , Anisoara Calinescu , Michael Wooldridge

An AI control protocol is a plan for usefully deploying AI systems that aims to prevent an AI from intentionally causing some unacceptable outcome. This paper investigates how well AI systems can generate and act on their own strategies for…

Machine Learning · Computer Science 2025-04-07 Alex Mallen , Charlie Griffin , Misha Wagner , Alessandro Abate , Buck Shlegeris

We sketch how developers of frontier AI systems could construct a structured rationale -- a 'safety case' -- that an AI system is unlikely to cause catastrophic outcomes through scheming. Scheming is a potential threat model where AI…

We present a quantitative model for tracking dangerous AI capabilities over time. Our goal is to help the policy and research community visualise how dangerous capability testing can give us an early warning about approaching AI risks. We…

Artificial Intelligence · Computer Science 2024-12-23 Paolo Bova , Alessandro Di Stefano , The Anh Han

Despite rapid growth in multimodal large language models (MLLMs), their reasoning traces remain opaque: it is often unclear which modality drives a prediction, how conflicts are resolved, or when one stream dominates. In this paper, we…

Artificial Intelligence · Computer Science 2025-11-05 Chenyu Zhang , Minsol Kim , Shohreh Ghorbani , Jingyao Wu , Rosalind Picard , Patricia Maes , Paul Pu Liang

As artificial intelligence (AI) models are scaled up, new capabilities can emerge unintentionally and unpredictably, some of which might be dangerous. In response, dangerous capabilities evaluations have emerged as a new risk assessment…

Computers and Society · Computer Science 2023-10-03 Jide Alaga , Jonas Schuett
‹ Prev 1 2 3 10 Next ›