Step 4: tool results

read_page

{
  "url": "https://arxiv.org/abs/2603.04426",
  "max_chars": 3000
}

{"ok":true,"url":"https://arxiv.org/abs/2603.04426","text":"[2603.04426] Delta-Crosscoder: Robust Crosscoder Model Diffing in Narrow Fine-Tuning Regimes\nSkip to main content\nSearch\nSubmit\nDonate\nLog in\nSearch arXiv\nPress Enter to search &middot; Advanced search\n-->\nComputer Science > Machine Learning\narXiv:2603.04426 (cs)\n[Submitted on 16 Feb 2026]\nTitle: Delta-Crosscoder: Robust Crosscoder Model Diffing in Narrow Fine-Tuning Regimes\nAuthors: Aly Kassem , Thomas Jiralerspong , Negar Rostamzadeh , Golnoosh Farnadi\nView a PDF of the paper titled Delta-Crosscoder: Robust Crosscoder Model Diffing in Narrow Fine-Tuning Regimes, by Aly Kassem and 3 other authors\nView PDF\nHTML (experimental)\nAbstract: Model diffing methods aim to identify how fine-tuning changes a model's internal representations. Crosscoders approach this by learning shared dictionaries of interpretable latent directions between base and fine-tuned models. However, existing formulations struggle with narrow fine-tuning, where behavioral changes are localized and asymmetric. We introduce Delta-Crosscoder, which combines BatchTopK sparsity with a delta-based loss prioritizing directions that change between models, plus an implicit contrastive signal from paired activations on matched inputs. Evaluated across 10 model organisms, including synthetic false facts, emergent misalignment, subliminal learning, and taboo word guessing (Gemma, LLaMA, Qwen; 1B-9B parameters), Delta-Crosscoder reliably isolates latent directions causally responsible for fine-tuned behaviors and enables effective mitigation, outperforming SAE-based baselines, while matching the Non-SAE-based. Our results demonstrate that crosscoders remain a powerful tool for model diffing.\nSubjects:\nMachine Learning (cs.LG) ; Artificial Intelligence (cs.AI)\nCite as:\narXiv:2603.04426 [cs.LG]\n(or\narXiv:2603.04426v1 [cs.LG] for this version)\nhttps://doi.org/10.48550/arXiv.2603.04426\nFocus to learn more\narXiv-issued DOI via DataCite\nSubmission history\nFrom: Thomas Jiralerspong [ view email ] [v1]\nMon, 16 Feb 2026 23:35:29 UTC (5,445 KB)\nFull-text links:\nAccess Paper:\nView a PDF of the paper titled Delta-Crosscoder: Robust Crosscoder Model Diffing in Narrow Fine-Tuning Regimes, by Aly Kassem and 3 other authors\nView PDF\nHTML (experimental)\nTeX Source\nview license\nAdditional Features\nAudio Summary\nCurrent browse context:\ncs.LG\n< prev\n|\nnext >\nnew\n|\nrecent\n| 2026-03\nChange to browse by:\ncs\ncs.AI\nReferences & Citations\nSciX\nGoogle Scholar\nSemantic Scholar\nexport BibTeX citation\nLoading...\nBibTeX formatted citation\n&times;\nloading...\nData provided by:\nBookmark\nBibliographic Tools\nBibliographic and Citation Tools\nBibliographic Explorer Toggle\nBibliographic Explorer ( What is the Explorer? )\nConnected Papers Toggle\nConnected Papers ( What is Connected Papers? )\nLitmaps Toggle\nLitmaps ( What is Litmaps? )\nscite.ai Toggle\nscite Smart Citations ( What are Smart Citations? )\nCode, Data, Media\nCode, Data and Media Associated with this Article\nalphaXiv Toggle\nalphaXiv ( What is alphaXiv? )\nLinks to Code Tog","chars":3000,"total_chars":4581,"offset":0,"truncated":true,"next_offset":3000}

read_page

{
  "url": "https://arxiv.org/abs/2602.22755",
  "max_chars": 3000
}

{"ok":true,"url":"https://arxiv.org/abs/2602.22755","text":"[2602.22755] AuditBench: Evaluating Alignment Auditing Techniques on Models with Hidden Behaviors\nSkip to main content\nSearch\nSubmit\nDonate\nLog in\nSearch arXiv\nPress Enter to search · Advanced search\n-->\nComputer Science > Computation and Language\narXiv:2602.22755 (cs)\n[Submitted on 26 Feb 2026 ( v1 ), last revised 1 Oct 2026 (this version, v4)]\nTitle: AuditBench: Evaluating Alignment Auditing Techniques on Models with Hidden Behaviors\nAuthors: Abhay Sheshadri , Aidan Ewart , Elias Kempf , Kai Fronsdal , Isha Gupta , Simon Schrodi , Arthur Conmy , Samuel R. Bowman , Sara Price , Samuel Marks , Rowan Wang\nView a PDF of the paper titled AuditBench: Evaluating Alignment Auditing Techniques on Models with Hidden Behaviors, by Abhay Sheshadri and 10 other authors\nView PDF\nHTML (experimental)\nAbstract: We introduce AuditBench, an alignment auditing benchmark. AuditBench consists of 56 language models with implanted hidden behaviors. Each model has one of 14 concerning behaviors--such as sycophantic deference, opposition to AI regulation, or secret geopolitical loyalties--which it does not confess to when directly asked. AuditBench models are highly diverse--some are subtle, while others are overt, and we use varying training techniques both for implanting behaviors and training models not to confess. To demonstrate AuditBench's utility, we develop an investigator agent that autonomously employs a configurable set of auditing tools. By measuring investigator agent success using different tools, we can evaluate their efficacy. Notably, we observe a tool-to-agent gap, where tools that perform well in standalone non-agentic evaluations fail to translate into improved performance when used with our investigator agent. We find that our most effective tools involve scaffolded calls to auxiliary models that generate diverse prompts for the target. White-box interpretability tools can be helpful, but the agent performs best with black-box tools. We also find that audit success varies greatly across training techniques: models trained on synthetic documents are easier to audit than models trained on demonstrations, with better adversarial training further increasing auditing difficulty. We release our models, agent, and evaluation framework to support future quantitative, iterative science on alignment auditing.\nSubjects:\nComputation and Language (cs.CL)\nCite as:\narXiv:2602.22755 [cs.CL]\n(or\narXiv:2602.22755v4 [cs.CL] for this version)\nhttps://doi.org/10.48550/arXiv.2602.22755\nFocus to learn more\narXiv-issued DOI via DataCite\nSubmission history\nFrom: Abhay Sheshadri [ view email ] [v1]\nThu, 26 Feb 2026 08:43:07 UTC (1,303 KB)\n[v2]\nTue, 3 Mar 2026 06:42:39 UTC (1,090 KB)\n[v3]\nMon, 9 Mar 2026 18:35:46 UTC (1,089 KB)\n[v4]\nThu, 1 Oct 2026 00:44:05 UTC (1,684 KB)\nFull-text links:\nAccess Paper:\nView a PDF of the paper titled AuditBench: Evaluating Alignment Auditing Techniques on Models with Hidden Behaviors, by Abhay Sheshadri and 10 other authors\nView PDF\nHTM","chars":3000,"total_chars":5381,"offset":0,"truncated":true,"next_offset":3000}