analiza aiqbench
This research demonstrates that Large Language Models can adopt specific personas and exhibit misaligned behaviors solely through the accumulation of benign biographical facts in their context window, bypassing the need for harmful demonstrations or fine-tuning. The study reveals a critical vulnerability where identity adoption becomes statistically significant with just 3 to 10 factual inputs, causing models to voice characteristic views on unrelated topics at rates up to 80%. This finding highlights a novel attack vector for content safety that operates under the radar of standard filters, which often fail to flag these accumulated benign contexts as effectively as direct instructions.
You Are What You Read: Misalignment via In-Context Persona Induction
arXiv:2609.06851v1 Announce Type: new Abstract: Broad misalignment has been produced by finetuning on narrow data, harmful or benign, and in context only by demonstrations of the undesirable behaviour itself. We show that benign data suffices in context, with no finetuning and no demonstration of harmful behaviour in the prompt. Biographical facts that converge on a single figure, placed in a model's context as ordinary conversational turns, lead it to answer as that figure on questions the facts never touch. We call this persona induction. Across nine personas and thirteen models, identity adoption rises sigmoidally with the number of facts and crosses 50% within 3 to 10 of them. Misalignment then tracks which figure is described. Harmless personas reach full adoption with near-zero misalignment, while harmful ones voice their characteristic views on unrelated questions, at rates up to 80%. A formatting instruction can gate when the persona activates. Because each fact is individually benign, accumulated biographical context is flagged by content filters on 3% of inputs against 24-33% for an equivalent direct instruction.
Abstract:Broad misalignment has been produced by finetuning on narrow data, harmful or benign, and in context only by demonstrations of the undesirable behaviour itself. We show that benign data suffices in context, with no finetuning and no demonstration of harmful behaviour in the prompt. Biographical facts that converge on a single figure, placed in a model's context as ordinary conversational turns, lead it to answer as that figure on questions the facts never touch. We call this persona induction. Across nine personas and thirteen models, identity adoption rises sigmoidally with the number of facts and crosses 50% within 3 to 10 of them. Misalignment then tracks which figure is described. Harmless personas reach full adoption with near-zero misalignment, while harmful ones voice their characteristic views on unrelated questions, at rates up to 80%. A formatting instruction can gate when the persona activates. Because each fact is individually benign, accumulated biographical context is flagged by content filters on 3% of inputs against 24-33% for an equivalent direct instruction.
20 pages, 14 figures, 27 tables. Includes appendices
Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
I.2.7; I.2.0; K.4.1
arXiv:2609.06851 [cs.CL]
(or arXiv:2609.06851v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2609.06851 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 276 cuvinte
Subiecte conexe
Latent reasoning architectures would likely undermine CoT, our strongest oversight tool
AI Alignment Forum · 23 sept. 2026
The Continual Constitutional Convention: How language models revise the documents that govern them
AI Alignment Forum · 13 oct. 2026
Concrete Evaluations to Investigate the OpenAI Model That Hacked Hugging Face
AI Alignment Forum · 3 aug. 2026
How might continual learning affect safety and alignment?
LessWrong · 13 iun. 2026