12 papers
Dense Is Not Enough: Hierarchical Supervision Allocation for Long-Horizon On-Policy Distillation
Yuhao Sun, Binrui Wu, Zhuoer Xu +5
On-policy distillation (OPD) transfers the capabilities of a large language model to a smaller student by providing teacher supervision on the student's own rollouts. In long-horiz…
HazardAuditor: From Executable Threats to Safer Computer-Use Agents
Yunhao Feng, Ruixiao Lin, Ming Wen +5
Computer-use agents increasingly interact with browsers, terminals, file systems, and external services, introducing safety risks that emerge through runtime behavior rather than g…
UI-Venus-2 Technical Report
Venus Team, Zhuohan Cai, Haoxing Chen +28
Multimodal GUI agents have emerged as a promising paradigm for digital task automation, yet transitioning from benchmark-oriented models to dependable real-world applications remai…
Safety Testing LLM Agents at Scale: From Risk Discovery to Evidence-Grounded Verification
Yunhao Feng, Ruixiao Lin, Ming Wen +12
LLM agents increasingly perform autonomous actions through external tools, leading to complex and evolving safety risks. However, existing safety testing targets expert-designed sa…
BraveGuard: From Open-World Threats to Safer Computer-Use Agents
Yunhao Feng, Xiaohu Du, Xinhao Deng +13
Computer-use agents extend language models from text generation to sustained interaction with files, terminals, browsers, and external tools. This shift creates safety risks that a…
Constitutional On-Policy Safe Distillation
Ming Wen, Yuxuan Liu, Kun Yang +8
On-policy self-distillation (OPSD) has emerged as an efficient post-training paradigm by using a teacher conditioned on privileged information to provide dense token-level supervis…
Rewriting the Response Path: Silent Tampering and Provider-Signed Defense in BYOK LLM Agents
Mingyu Luo, Zihan Zhang, Zesen Liu +7
LLM agents convert model outputs into consequential actions, including communications, code changes, and financial transactions. Developers often trust evidence such as test result…
OOD-MMSafe: Advancing MLLM Safety from Harmful Intent to Hidden Consequences
Ming Wen, Kun Yang, Jingyu Zhang +4
While safety alignment for Multimodal Large Language Models (MLLMs) has gained significant attention, current paradigms primarily target malicious intent or situational violations.…
Autoregressive, Yet Revisable: In Decoding Revision for Secure Code Generation
Chengran Yang, Zichao Wei, Heminghao Deng +6
Large Language Model (LLM) based code generation is predominantly formulated as a strictly monotonic process, appending tokens linearly to an immutable prefix. This formulation con…
A Safety Report on GPT-5.2, Gemini 3 Pro, Qwen3-VL, Grok 4.1 Fast, Nano Banana Pro, and Seedream 4.5
Xingjun Ma, Yixu Wang, Hengyuan Xu +18
The rapid evolution of Large Language Models (LLMs) and Multimodal Large Language Models (MLLMs) has driven major gains in reasoning, perception, and generation across language and…
PATCHEVAL: A New Benchmark for Evaluating LLMs on Patching Real-World Vulnerabilities
Zichao Wei, Jun Zeng, Ming Wen +8
Software vulnerabilities are increasing at an alarming rate. However, manual patching is both time-consuming and resource-intensive, while existing automated vulnerability repair (…
An Empirical Study of Production Incidents in Generative AI Cloud Services
Haoran Yan, Yinfang Chen, Minghua Ma +10
The ever-increasing demand for generative artificial intelligence (GenAI) has motivated cloud-based GenAI services such as Azure OpenAI Service and Amazon Bedrock. Like any large-s…