SaferAI

This paper examines what rigorous, quantitative risk modeling for advanced AI should look like in practice — reviewing five research traditions, comparing two leading approaches, and drawing on a 22-expert workshop to set out an agenda of open questions and priorities for the fie

SaferAI's independent evaluation of GLM-5.2, the first in Europe, tests Zhipu AI's open-weight flagship across the four systemic risk areas in the EU Code of Practice and finds frontier-level capability on cyber and biology benchmarks without the safeguards frontier developers ap

Advanced AI systems offer substantial benefits but also introduce risks. In 2025, AI-enabled cyber offense has emerged as a concrete example. This technical report applies a quantitative risk modeling methodology (described in full in a companion paper) to this domain. We develop

Although general-purpose AI systems offer transformational opportunities in science and industry, they simultaneously raise critical concerns about safety, misuse, and potential loss of control. Despite these risks, methods for assessing and managing them remain underdeveloped. E

In this research memo, the Oxford Martin AIGI brings together diverse stakeholders to sketch out the contours of a “gold-standard” framework for risk tiering—balancing quantitative and qualitative assessments, lifecycle classification, and the integration of benefit-risk reasonin

This report contains an analysis of the commonalities and differences between the G7 Hiroshima Process International Code of Conduct for Organizations Developing Advanced AI Systems and the EU AI Act text on general-purpose AI models. There is substantial commonality between the

The literature and multiple experts point to many potential risks from large language models (LLMs), but there are still very few direct measurements of the actual harms posed. AI risk assessment has so far focused on measuring the models' capabilities, but the capabilities of mo

SaferAI published a paper summarizing findings from a workshop focused on what rigorous, quantitative AI risk modeling should look like in practice and an agenda of open questions for the field.

SaferAI published commentary arguing that governance needs improvements to the evaluation pipeline and discusses how evidence fails to translate into preventative action; the post includes an editor’s update describing changes relative to initial publication timing.

SaferAI released an independent evaluation of GLM-5.2, describing testing across the four systemic risk areas in the EU General-Purpose AI Code of Practice and reporting findings about frontier capability relative to safeguards.

SaferAI published an analysis of frontier AI safety frameworks across multiple companies, describing an overall assessment gap and best-practice areas that SaferAI characterizes as currently strongest in the industry.

SaferAI published a position memo arguing for investment in high-risk, high-reward AI reliability research and laying out research directions it considers important for making advanced AI reliable and safe.