Dawn Song
Professor
Studies AI safety and security, agentic AI and decentralized technology.
Dawn Song's lab at UC Berkeley works on trustworthy & safe ML. 112 papers since 2023, 78 at top venues, mostly ICLR and NeurIPS. 12 alumni. Applications close .
At a glance
- papers, 2023–2026
- 29/ year
- at T1 venues
- 70%
- PhD students
- —
- MS students
- —
- alumni listed
- 12
Papers by yearfrom DBLP, 2023–2026
- 23 papers2026 · T1 19 · collecting
- 32 papers2025 · T1 23
- 34 papers2024 · T1 21
- 23 papers2023 · T1 15
T1 venuesOtherVenues: ICLR, NeurIPS, ICML, USENIX Security, ACL
Research focuspaper titles that use each topic's words
- safety6 papers
- decentralized3 papers
- Trustworthy & safe ML1 paper
- agentic0 papers
- computer sciences0 papers
Paper list
From DBLP.
202623 papers
- SACMATA Critical Evaluation of Defenses against Prompt Injection Attacks: [Dataset/Tool Paper] (opens in a new tab)
- KDD (T1)A Hands-on Tutorial for Diffusion Language Models with dLLM (opens in a new tab)
- ICLR (T1)AgentSynth: Scalable Task Generation for Generalist Computer-Use Agents (opens in a new tab)
- AAAI (T1)Can Editing LLMs Inject Harm? (opens in a new tab)
- ICLR (T1)Can Small Training Runs Reliably Guide Data Curation? Rethinking Proxy-Model Practice (opens in a new tab)
- KDD (T1)CTIConnect: A Benchmark for Retrieval-Augmented LLMs over Heterogeneous Cyber Threat Intelligence (opens in a new tab)
- ICLR (T1)CyberGym: Evaluating AI Agents' Real-World Cybersecurity Capabilities at Scale (opens in a new tab)
- ICLR (T1)DevOps-Gym: Benchmarking AI Agents in Software DevOps Cycle (opens in a new tab)
- ACL (T1)dLLM: Simple Diffusion Language Modeling (opens in a new tab)
- Findings of ACLFico: Evaluating Vision-Language Models under Visual Fidelity and Compression at Scale (opens in a new tab)
- Commun. ACMFormal Reasoning Meets LLMs: Toward AI for Mathematics and Verification (opens in a new tab)
- ICLR (T1)Holistic Agent Leaderboard: The Missing Infrastructure for AI Agent Evaluation (opens in a new tab)
202532 papers
- NeurIPS (T1)A Sustainable AI Economy Needs Data Deals That Work for Generators (opens in a new tab)
- Findings of EMNLPAGENTVIGIL: Automatic Black-Box Red-teaming for Indirect Prompt Injection against LLM Agents (opens in a new tab)
- ICLR (T1)AIR-BENCH 2024: A Safety Benchmark based on Regulation and Policies Specified Risk Categories (opens in a new tab)
- AAAI/ACM AIESAligning AI with Public Values: Deliberation and Decision-Making for Governing Multimodal LLMs in Political Video Analysis (opens in a new tab)
- ICLR (T1)An Undetectable Watermark for Generative Image Models (opens in a new tab)
- NeurIPS (T1)BountyBench: Dollar Impact of AI Agent Attackers and Defenders on Real-World Cybersecurity Systems (opens in a new tab)
- ICLR (T1)Capturing the Temporal Dependence of Training Data Influence (opens in a new tab)
- AAAI (T1)CodeHalu: Investigating Code Hallucinations in LLMs via Execution-based Verification (opens in a new tab)
- Findings of ACLCOSMIC: Generalized Refusal Direction Identification in LLM Activations (opens in a new tab)
- EuroS&PCTINexus: Automatic Cyber Threat Intelligence Knowledge Graph Construction Using Large Language Models (opens in a new tab)
- ICLR (T1)Data Shapley in One Training Run (opens in a new tab)
- SPDataSentinel: A Game-Theoretic Detection of Prompt Injection Attacks (opens in a new tab)
202434 papers
- CHI (T1)"I Can't Believe It's Not Custodial!": Usable Trustless Decentralized Key Management (opens in a new tab)
- ICML (T1)Agent Instructs Large Language Models to be General Zero-Shot Reasoners (opens in a new tab)
- NeurIPS (T1)AgentPoison: Red-teaming LLM Agents via Poisoning Memory or Knowledge Bases (opens in a new tab)
- EMNLP (T1)BEEAR: Embedding-based Adversarial Removal of Safety Backdoors in Instruction-tuned Language Models (opens in a new tab)
- IEEE Trans. Vis. Comput. Graph.Berkeley Open Extended Reality Recordings 2023 (BOXRR-23): 4.7 Million Motion Capture Recordings from 105,000 XR Users (opens in a new tab)
- NeurIPS (T1)Boosting Alignment for Post-Unlearning Text-to-Image Generative Models (opens in a new tab)
- ICML (T1)C-RAG: Certified Generation Risks for Retrieval-Augmented Language Models (opens in a new tab)
- NeurIPS (T1)Data Free Backdoor Attacks (opens in a new tab)
- ICML (T1)Decoding Compressed Trust: Scrutinizing the Trustworthiness of Efficient LLMs Under Compression (opens in a new tab)
- VR WorkshopDeep Motion Masking for Secure, Usable, and Scalable Real-Time Anonymization of Ecological Virtual Reality Motion Data (opens in a new tab)
- ICLR (T1)Effective and Efficient Federated Tree Learning on Hybrid Data (opens in a new tab)
- ICML (T1)GRATH: Gradual Self-Truthifying for Large Language Models (opens in a new tab)
202323 papers
- ICML (T1)Adversarial Collaborative Learning on Non-IID Features (opens in a new tab)
- NeurIPS (T1)BIRD: Generalizable Backdoor Detection and Removal for Deep Reinforcement Learning (opens in a new tab)
- IACR Cryptol. ePrint Arch.Blockchain Large Language Models (opens in a new tab)
- AISTATS (T1)Byzantine-Robust Federated Learning with Optimal Statistical Rates (opens in a new tab)
- TPAMI (T1)Dataset Security for Machine Learning: Data Poisoning, Backdoor Attacks, and Defenses (opens in a new tab)
- ICBCDecentralizing Custodial Wallets with MFKDF (opens in a new tab)
- NeurIPS (T1)DecodingTrust: A Comprehensive Assessment of Trustworthiness in GPT Models (opens in a new tab)
- ICLR (T1)DensePure: Understanding Diffusion Models for Adversarial Robustness (opens in a new tab)
- NeurIPS (T1)DiffAttack: Evasion Attacks Against Diffusion-Based Adversarial Purification (opens in a new tab)
- Proc. Priv. Enhancing Technol.Exploring the Privacy Risks of Adversarial VR Game Design (opens in a new tab)
- UIST (T1)Going Incognito in the Metaverse: Achieving Theoretically Optimal Privacy-Usability Tradeoffs in VR (opens in a new tab)
- Found. Trends Priv. Secur.Identifying and Mitigating the Security Risks of Generative AI (opens in a new tab)
Where alumni went12 listed
Apply to UC BerkeleyFall 2027
UC Berkeley EECS Computer Science PhD
- Round
- Single round
- Opens
- —
- GRE
- not required
- English
- TOEFL/IELTS (non-English-country degrees)
- Fee
- See program page
Confirmed on the program page.
Similar labsnearest by research summary
- USUIUC
Richard Y. Zhang
- Learning theory & optimization
- CNTsinghua
Kaifeng Lyu
- Learning theory & optimization
- USMichigan
Atul Prakash
- Trustworthy & safe ML
- DETUM
Stefanie Jegelka
- Learning theory & optimization
- SGNTU
Yeshwanth Cherapanamjeri
- Learning theory & optimization
- USCMU
Matt Fredrikson
- Trustworthy & safe ML
Your notes
Save this lab to keep notes.
Data
Some numbers could not be verified:
- The member list could not be found on the lab site.
- Collected
- Verified
- Papers
- DBLP
- Members
- Not collected
- Alumni
- Lab homepage
Tiers are venue classifications, not lab rankings.
UC Berkeleycloses