Future Events as Backdoor Triggers: Investigating Temporal Vulnerabilities in LLMs

AmazUtah_NLP at SemEval-2024 Task 9: A MultiChoice Question Answering System for Commonsense Defying Reasoning


View a PDF of the paper titled Future Events as Backdoor Triggers: Investigating Temporal Vulnerabilities in LLMs, by Sara Price and 3 other authors

View PDF
HTML (experimental)

Abstract:Backdoors are hidden behaviors that are only triggered once an AI system has been deployed. Bad actors looking to create successful backdoors must design them to avoid activation during training and evaluation. Since data used in these stages often only contains information about events that have already occurred, a component of a simple backdoor trigger could be a model recognizing data that is in the future relative to when it was trained. Through prompting experiments and by probing internal activations, we show that current large language models (LLMs) can distinguish past from future events, with probes on model activations achieving 90% accuracy. We train models with backdoors triggered by a temporal distributional shift; they activate when the model is exposed to news headlines beyond their training cut-off dates. Fine-tuning on helpful, harmless and honest (HHH) data does not work well for removing simpler backdoor triggers but is effective on our backdoored models, although this distinction is smaller for the larger-scale model we tested. We also find that an activation-steering vector representing a model’s internal representation of the date influences the rate of backdoor activation. We take these results as initial evidence that, at least for models at the modest scale we test, standard safety measures are enough to remove these backdoors.

Submission history

From: Sara Price [view email]
[v1]
Thu, 4 Jul 2024 18:24:09 UTC (4,621 KB)
[v2]
Wed, 17 Jul 2024 18:45:46 UTC (4,621 KB)
[v3]
Mon, 23 Dec 2024 19:24:44 UTC (5,030 KB)



Source link
lol

By stp2y

Leave a Reply

Your email address will not be published. Required fields are marked *

No widgets found. Go to Widget page and add the widget in Offcanvas Sidebar Widget Area.