We examine why Mythos 5 pursued an unsanctioned supply-chain attack and identify broader learnings about model behaviour and oversight.

Specialised biological AI models (or tools) are driving rapid advances in biotechnology. They are typically released as open-weight models and are smaller than frontier large language models (LLMs), ...

Shiv Muthupandiyan, Tolga H. Dur, Mac Walker, Timo Flesch, Mahmoud Ghanem, Steph Suddell, Giles Harper-Donnelly, Sophie Rose, Jasper Götting, Coleman Breen, Jake Pencharz Large language models are ...

As AI systems are deployed in high-stakes environments, it becomes increasingly important that they act as their operators intend. Existing red-teaming efforts have demonstrated situations where AI ...

How do we test AI systems for dangerous capabilities without risking real-world harm? The more capable models become, the harder it is to safely evaluate them. When an agent can execute arbitrary code ...

Existing R&D evaluations may be a promising start We provided the participants with several example tasks from AI R&D evaluations that are currently under development, covering experiment ...

AI software engineering agents are increasingly being deployed to write production code. Unlike standard chatbots, these agents are often given tools and permissions to take actions such as exploring ...

Today, we’re releasing results from our work with Anthropic and the Alan Turing Institute to produce the largest study of data poisoning to date. Data poisoning occurs when individuals distribute ...

Building on the momentum of the Inspect toolkit, we’re open-sourcing parts of the research stack behind AISI's evaluations. In May 2024, the AI Security Institute open-sourced Inspect AI, a toolkit ...

Several leading AI companies have the stated goal of building AI systems that match or surpass humans across most cognitive domains. Such systems could have transformative effects, such as increasing ...

Standard evaluations cap how much compute AI agents can use. We show that raising those caps changes measured capability, the difficulty of tasks agents can solve, and how fast the frontier appears to ...