Hair-Trigger Alignment: Black-Box Evaluation Cannot Guarantee Post-Update Alignment
Fuente:
arXiv
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866914293096644608 |
|---|---|
| author | Bakman, Yavuz Yaldiz, Duygu Nur Avestimehr, Salman Karimireddy, Sai Praneeth |
| author_facet | Bakman, Yavuz Yaldiz, Duygu Nur Avestimehr, Salman Karimireddy, Sai Praneeth |
| contents | Large Language Models (LLMs) are rarely static and are frequently updated in practice. A growing body of alignment research has shown that models initially deemed "aligned" can exhibit misaligned behavior after fine-tuning, such as forgetting jailbreak safety features or re-surfacing knowledge that was intended to be forgotten. These works typically assume that the initial model is aligned based on static black-box evaluation, i.e., the absence of undesired responses to a fixed set of queries. In contrast, we formalize model alignment in both the static and post-update settings and uncover a fundamental limitation of black-box evaluation. We theoretically show that, due to overparameterization, static alignment provides no guarantee of post-update alignment for any update dataset. Moreover, we prove that static black-box probing cannot distinguish a model that is genuinely post-update robust from one that conceals an arbitrary amount of adversarial behavior which can be activated by even a single benign gradient update. We further validate these findings empirically in LLMs across three core alignment domains: privacy, jailbreak safety, and behavioral honesty. We demonstrate the existence of LLMs that pass all standard black-box alignment tests, yet become severely misaligned after a single benign update. Finally, we show that the capacity to hide such latent adversarial behavior increases with model scale, confirming our theoretical prediction that post-update misalignment grows with the number of parameters. Together, our results highlight the inadequacy of static evaluation protocols and emphasize the urgent need for post-update-robust alignment evaluation. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_22313 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Hair-Trigger Alignment: Black-Box Evaluation Cannot Guarantee Post-Update Alignment Bakman, Yavuz Yaldiz, Duygu Nur Avestimehr, Salman Karimireddy, Sai Praneeth Machine Learning Large Language Models (LLMs) are rarely static and are frequently updated in practice. A growing body of alignment research has shown that models initially deemed "aligned" can exhibit misaligned behavior after fine-tuning, such as forgetting jailbreak safety features or re-surfacing knowledge that was intended to be forgotten. These works typically assume that the initial model is aligned based on static black-box evaluation, i.e., the absence of undesired responses to a fixed set of queries. In contrast, we formalize model alignment in both the static and post-update settings and uncover a fundamental limitation of black-box evaluation. We theoretically show that, due to overparameterization, static alignment provides no guarantee of post-update alignment for any update dataset. Moreover, we prove that static black-box probing cannot distinguish a model that is genuinely post-update robust from one that conceals an arbitrary amount of adversarial behavior which can be activated by even a single benign gradient update. We further validate these findings empirically in LLMs across three core alignment domains: privacy, jailbreak safety, and behavioral honesty. We demonstrate the existence of LLMs that pass all standard black-box alignment tests, yet become severely misaligned after a single benign update. Finally, we show that the capacity to hide such latent adversarial behavior increases with model scale, confirming our theoretical prediction that post-update misalignment grows with the number of parameters. Together, our results highlight the inadequacy of static evaluation protocols and emphasize the urgent need for post-update-robust alignment evaluation. |
| title | Hair-Trigger Alignment: Black-Box Evaluation Cannot Guarantee Post-Update Alignment |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2601.22313 |