Post

Announcing: the #SaTML2027 competitions and workshops! We have an amazing lineup of two competitions and four workshops. The workshops will be on May 3, ahead of the conference that runs May 3-6 in Reykjavik, Iceland. Read on to learn more about the competitions and workshops!

#SaTML2027 will have two competitions: 1. LatentArena: tests representation-level safety monitors. An ensemble of monitors covers a variety of harmful-behaviour categories. Two phases, first infer monitor behaviour, then try to evade it. Website forthcoming.

2. Ravensport: How much warning can ML give before a cyberattack? Can an agent mitigate attacks without disrupting operations? Participants monitor and defend a realistic infrastructure as multi-stage attacks unfold. github.com/alan-turing-...

#SaTML2027 also has four workshops: 1. New AdvML-Frontiers: Human Centered Trustworthy Machine Learning. Focused on understanding which risks lead to real-world harm or which protections are truly needed. advml-frontier.github.io/satml2027/

2. MIRROR: Meta-science in AI Security Research. How can our community conduct more reliable & reproducible research? Looking in 3 dir: backwards (where past research goes wrong), forwards (good research practices), inwards (how do LLMs affect research?) mirror-workshop.io

4. TrustTFM: Trustworthiness of Tabular Foundation Models (TFMs). Tabular data is ubiquitous, and regulations make trustworthiness essential: robustness, security, privacy, explainability, fairness, and compliance risks. karimtito.github.io/trusttfm-2027/

More details on the website: satml.org/competitions/ satml.org/workshops/ Please check them out! We look forward to your participation in the competitions, workshops, and the main program itself.