June brought another milestone for Project Ace. Our autonomous table tennis robot, Ace, beat ranked professional players under official competition rules. We also presented six papers at CVPR 2026 in Denver. And our AI ethics research reached a wider audience through a TIME op-ed on data rights. Here is a look at what we worked on this month, and what is coming next.
What Happened After the Nature Paper: Ace vs. Professional Players
In April, we published Ace in Nature. Ace is our autonomous table tennis robot. The paper showed that Ace could compete with and beat elite players, but it could not yet outplay professionals. This month we shared what happened next.
Between February and April 2026, Ace won matches against seven ranked professionals under official competition rules. The run included a win over Miyuu Kihara, currently World No. 26 in women’s singles and the highest-ranked player Ace has beaten to date. Ace also beat two-time Olympic silver medalist Miu Hirano.
After the match, Hirano described the experience in one line: “It’s really strong. Is there really anyone who can beat this?”
The team reached this level mostly through retraining rather than redesign. The improvements spanned control, simulation, hardware, and perception. Ace now uses a single reinforcement learning policy that performs nine distinct skills, serves from both sides of the court, and reacts faster than before. The result is a system that moves with greater speed, returns more consistently, and wins at a professionally ranked level.
The work drew broad coverage, with outlets including Robotics and Automation News, Technology Magazine, and ScienceAlert reporting on Ace and what it means for physical AI.
Sony AI and its collaborators presented six papers at CVPR 2026 in Denver, held June 3 to 7. The work spanned generative modeling, 3D scene understanding, video-to-audio synthesis, domain-adaptive perception, and visual token efficiency. Each paper targets a different constraint on building systems that hold up outside the lab.
Peter Stone, Chief Scientist at Sony AI, gave a keynote at the first Workshop on Deployment of Foundation Models for Embodied AI. His talk examined how large foundation models can be deployed in autonomous systems, from self-driving cars to legged robots.
Chieh-Hsin Lai and Yuki Mitsufuji taught a tutorial on diffusion and flow-map models for fast sampling. It was grounded in their textbook, The Principles of Diffusion Models. Slides and recordings will be available after the conference.
The ACM Conference on Fairness, Accountability, and Transparency ran June 25 to 28 in Montréal. FAccT is the leading venue for research on fairness and accountability in AI. Alice Xiang, Global Head of AI Governance at Sony, serves on the conference steering committee. This year, Sony AI researchers and collaborators contribute through a tutorial and three CRAFT sessions. Here is a brief look at each.
Tutorial: Counterfactual fairness analysis in language-vision models
Authors: Kathleen C. Fraser, Phillip Howard, Jieyu Zhao, Margaret McKay, Morgan Klaus Scheuerman
Counterfactual fairness holds that changing a social attribute in a model’s input should not change its output. This tutorial asks how to develop and use counterfactual image datasets to assess bias in large vision-language models. It covers dataset creation, the tradeoffs of collecting versus generating images, and the axes of discrimination the literature has studied and overlooked. It ends with a hands-on session so attendees can test the methodology themselves.
CRAFT: Actualizing Ethical Principles for Curating Large-Scale Training Datasets in the Era of Massive AI Models
Authors: Silvia Cazacu, Alice Qian, Dora Zhao, Kathleen Pine, Shawn Walker, Hong Shen, Laura Dabbish, Georgia Panagiotidou, Morgan Klaus Scheuerman
AI depends on huge datasets shaped by human decisions, labor, and power structures. This session questions the assumptions behind large-scale dataset curation across three areas: composition, process, and release. The goal is a shared framework for fairer data practices that treats data work as a collective responsibility, not just a technical task.
CRAFT: Co-designing Responsible and Ethical Disclosure of AI in Tools (CREDIT)
Authors: Hyo Jin Do, Molly Q Feldman, Finola Finn, Jessica He, Angel Hsing-Chi Hwang, Donal Khosrowi, Seyun Kim, Advait Sarkar, Morgan Klaus Scheuerman
As generative AI becomes common at work, so does the need to disclose its use. This workshop brings together three perspectives, users, practitioners, and policy experts, to co-design practical disclosure approaches. The aim is to maintain trust and clear communication, and to produce a repository of thoughtfully designed disclosures.
CRAFT: Common Ground: Closing the Gap Between Algorithmic Collective Action Research and LGBTQIA2S+ Community Needs
Authors: Ulrich Aïvodji, Mina Alfaghih, Meghana Bhange, Elliot Creager, Jacob Hobbs, Michelle Lin, Yanan Long, Sarah Mathew, Jennifer Mickel, Annie Pullen Sansfaçon, Ruchira Ray, Morgan Klaus Scheuerman, Arjun Subramonian, Nandhini Swaminathan, Sabine Weber
This interactive workshop looks at how LGBTQIA2S+ people experience harm from algorithms and how they respond. Through scenario-based, small group discussions, participants identify types of harm and unmet needs across recommendations, moderation, and access to services. The goal is community-centered ideas for collective action that connect research with real community needs. Learn more about FAccT 2026: facctconference.org
To learn more about some of our research presented at FaCCT over the years, dive into the following blogs:
Marc Ferras, Staff AI Engineer, and Hakim Missoum, Strategy & Partnerships Manager at Sony AI, sat down with Dean Takahashi at GamesBeat to discuss Woosh, our foundation AI model built specifically for sound effect generation.
In the interview, Ferras notes, “The idea is to really help creative game developers make better games at a faster pace. It’s to empower creators and not to replace them.” Missoum also shared where Sony AI would like to see this technology used, “We would like to see this technology being used by audio teams across film studios and game studios, and see really how we could accelerate the workflows and get feedback from them to make their jobs easier. At the same time, also, what I would like to see is how can we use this technology to enable new ways of creating new experiences.”
Read the full interview here: Sony AI releases Woosh foundation model for sound effect generation | exclusive interview
Alice Xiang, Global Head of AI Governance at Sony and lead research scientist at Sony AI, wrote an op-ed for TIME. She argues that people have come to feel their data means everything to AI developers and almost nothing to themselves. She calls this data nihilism.
The piece makes the case for a third path between sacrificing data rights and falling behind in AI. Xiang calls it ethical innovation, built on consent and fair compensation. She points to Sony AI research as proof that high-quality datasets can be sourced responsibly, including FHIBE, the first publicly available, consensually collected fairness benchmark for computer vision.
Read the op-ed in TIME: Are We Entering the Age of Data Nihilism?