Risks and Security

Note: this is a hybrid meeting. We will be in room G.07 in the Informatics Forum for those attending in person.

Overview

Following a brief hiatus in July due to holidays, we’re back with two exciting speakers for the August meetup. This meetup will focus on agentic coding risks and security. We will hear from Philippe Charrière from Docker about running agents offline, on your own laptop, and Joseph Shingleton from the University of Glasgow will discuss insights on the impacts of AI in research software engineering.

In addition to presenting at the meetup, Philippe and colleagues from Docker will be running two workshops on Docker Sandboxes on Tuesday 25 August. You can book a place on either workshop.

Speakers

Philippe Charrière (Docker)

• Title: A (Tiny) Coding Agent on Your Laptop, Offline: Docker Agent + Docker Model Runner, Step by Step
• Abstract: Everyone is wiring up code agents, usually with a cloud API, a pile of glue code, and your source shipped off to a third party. This talk takes the opposite bet: we build a working (small) code agent without writing a single line of application code, running entirely on a small local model on your laptop. The whole agent is declared in YAML with docker-agent, and the model is served locally by Docker Model Runner behind an OpenAI-compatible API, no API key, no cloud, no bill … no vibe coding (I will use a 12b model) . Step by step, we go from a one-tool agent to a small team of specialised agents, and along the way you’ll understand exactly what an “agent loop” is and why it’s simpler than it looks.

Joseph Shingleton (University of Glasgow)

• Title: Metascience, AI and Research Software Engineering
• Abstract: AI is fundamentally changing the economics of research software production. On the one hand, AI promises to increase researchers’ productivity, reduce barriers to entry for computational research, and make our code more reproducible, reliable and efficient. On the other hand, it poses serious risks to the integrity and sustainability of our work, and to the professional identity of many practitioners. This talk will introduce three ongoing projects looking at how AI is changing the way we write research code: A community developed risk-register detailing the nuanced and specific challenges researchers have faced in AI adoption, a study of how AI can be used to support reproduction of computational findings, and a mixed-methods study on how we establish trust in the research code we write (or, indeed, generate).