Background & Biography
Dawn Song’s work began in computer security before her involvement in blockchain entrepreneurship. The MacArthur Foundation records undergraduate study at Tsinghua University, a master’s degree at Carnegie Mellon University and a doctorate at UC Berkeley in 2002. She taught at Carnegie Mellon from 2002 to 2007 before joining Berkeley’s faculty. Her 2010 MacArthur Fellowship recognized research into the interactions among software, hardware and networks that can create broad classes of security weaknesses.
Berkeley’s current faculty pages identify Song as a professor of computer science and co-director of the Center for Responsible, Decentralized Intelligence, usually called RDI. The center’s people page also identifies Christine Parlour as a co-director. Song’s listed research interests include security and privacy, decentralized technology and AI safety. Her academic work is based in the United States.
Her commercial activities have developed alongside that academic career. Contemporary reporting identifies her as a founder of Oasis Labs, established around the problem of using data while protecting it. She later co-founded Virtue AI in 2024. In her own June 2026 announcement, Song said that she and many members of the Virtue AI team would join Meta Superintelligence Labs, where she would serve as Vice President of AI Research. Berkeley continues to list her as a professor and RDI co-director alongside her new industry role.
Notable Contributions
Song’s early security work provides context for her later approach to decentralized systems. Berkeley’s account of her 2020 ACM SIGSAC Outstanding Innovation Award highlights contributions to dynamic taint analysis, vulnerability discovery and malware detection. It also describes BitBlaze, a platform for analyzing binary programs. The common problem is how to reason systematically about a program’s behavior, rather than relying only on a patch for each newly discovered failure.
Her record includes collaborative work on cryptographic protection of data. Berkeley’s announcement of the 2020 IEEE Symposium on Security and Privacy Test-of-Time Awards identifies three honored papers co-authored by Song, including Practical Techniques for Searches on Encrypted Data. That history matters for digital finance because confidentiality and useful computation are often in tension. A system may need to demonstrate or compute something about information without exposing every underlying record to every participant.
Ekiden offers a direct connection to blockchain infrastructure. Song is one of nine authors of the paper first submitted in 2018 and subsequently associated with the 2019 IEEE European Symposium on Security and Privacy. The work proposes combining a blockchain with trusted execution environments, separating consensus from computation to support confidential smart-contract execution. It addresses a limitation of public execution environments: applications can inherit the transparency of the ledger even when their users require confidential inputs.
The Ekiden paper also examines security problems introduced by combining those components. This is an important qualification. A trusted execution environment introduces its own assumptions, and a hybrid architecture must account for how attacks on one component affect another. The paper’s prototype results belong to its stated experimental setting; they are not a current performance guarantee for every Oasis application. Song’s contribution is collaborative research, not sole authorship of the underlying techniques.
Oasis Labs translated related privacy concerns into an entrepreneurial project. WIRED’s July 2018 reporting independently documented Song’s role and the company’s announced $45 million financing. Its account described ambitions to combine blockchain-based assurances with privacy-preserving computation. That historical coverage establishes the company’s origins and technical direction without proving that every proposed application reached production or that a funding announcement measured later adoption.
The company’s work also extends beyond blockchain-native applications. Oasis Labs describes a collaboration with Meta to measure potential differences in AI-model behavior using protected demographic information. Its published account identifies multiparty computation and other privacy technologies as components of the approach. This is evidence of an organizational project and its stated design; it should not be converted into a claim that Song personally implemented every component or that the system provides an unconditional guarantee against all information leakage.
A complementary strand of Song’s research examines whether privacy techniques can be used effectively in practice. She co-authored Lessons Learned: Surveying the Practicality of Differential Privacy in the Industry, first posted in 2022. The paper reviews industrial applications and the difficulties of translating theoretical privacy properties into working systems. For digital-asset developers, the relevance is methodological: specifying a privacy technique is only the beginning of explaining how data access, configuration and operational constraints affect the protection users actually receive.
Her current public work also addresses the security of AI agents. In a 2026 Berkeley interview, Song discussed prompt injection, accumulated errors in multistep tasks, evaluation and human oversight. She connected safer deployment with advances in system architecture and rigorous testing. RDI’s courses and events provide a channel for sharing that research with a wider audience. These activities broaden her professional scope while maintaining continuity with her longstanding interest in systems that can act on information without violating users’ expectations.
Why Katharos.ai Follows Dawn Song
Katharos.ai follows Song because privacy and security affect what decentralized applications can responsibly do. Financial or identity-related systems cannot be judged only by whether a transaction is recorded successfully. Developers also need to explain who can observe the underlying information, which components must be trusted and what happens when those assumptions fail. Song’s work offers concrete research and institutional activity against which those questions can be examined.
Her career also makes the route from academic work to commercial infrastructure visible. Papers, prototypes, company systems and educational programs are different contributions with different standards of evidence. Following them together is useful, provided that a scientific result is not stretched into a blanket product assurance and a founder’s reputation is not used as a substitute for independent technical assessment.
Areas to Follow
- Confidential-computing designs and the security assumptions of hardware-backed smart-contract execution. - Independent evaluation of privacy-preserving systems, including realistic threat models and operational limitations. - Oasis Labs’ published work and its relationship to applications using the Oasis network. - Differential privacy and other approaches that allow useful analysis while limiting disclosure. - Song’s AI-security research, Meta role and Berkeley teaching, especially where autonomous software interacts with sensitive systems.
Reporting Scope
This profile covers Song’s verified professional roles, selected co-authored research and documented entrepreneurial activity. Research proposals, experimental results and product descriptions are kept distinct. The profile does not attribute all work by Berkeley, Meta, Oasis Labs or the Oasis ecosystem to her personally, and it makes no claim about token performance, private holdings or investment suitability.
Areas of Influence
Song’s principal influence channels are security research, decentralized infrastructure, privacy engineering, entrepreneurship and technical education. Her work connects the protection of information with the systems that process and act on it. The relevant measure is the quality and practical consequences of specific contributions, including the limits documented by their authors, rather than an assumption that privacy or decentralization is achieved simply by adopting a particular label.
Katharos.ai publishes this profile for information, based on the sources listed. Inclusion is not an endorsement of the individual, their organizations, or any investment product.
Source notice: Information and images are attributed to the external sources listed. Katharos.ai does not endorse the subjects or claims described and does not independently certify the authenticity of every source statement or image.
Photo attribution: Dawn Song at ICLR 2025 (25 Apr 2025), Xuthoria, via Wikimedia Commons (https://commons.wikimedia.org/wiki/File:ICLR_2025_-_Dawn_Song_01.jpg). Cropped and resized; original and derivative licensed CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/).