Ken Miyachi
Co-founder and CEO, BitMind
BitMind, co-founded by Ken Jon Miyachi, trains deepfake detectors through an open adversarial competition on Bittensor Subnet 34, on the premise that a detector trained once falls behind the generators it is meant to catch.
Why the room needs them.
Ken Jon Miyachi, co-founder and CEO of BitMind, builds deepfake detection on Bittensor Subnet 34 and spoke at Proof of Talk Paris 2026. The appearances were a Bittensor Track subnet keynote on 2 June and the TensorFi side event on 3 June.
The full account.
Ken Jon Miyachi is co-founder and chief executive of BitMind, a company that builds deepfake detection on Bittensor Subnet 34. Crypto Briefing named Miyachi co-founder and CEO in an article of 27 July 2026, and the company's own team page gives the same title. Proof of Talk billed Miyachi as Founder, BitMind.
BitMind's team page describes Miyachi as a former senior tech lead at NEAR Foundation, the founder of LedgerSafe, and previously an Amazon engineer with expertise in large-scale distributed infrastructure. Author biographies published in 2025 add Polymer Labs to that list, along with academic research publications on blockchain from work at the San Diego Supercomputer Center. Crowdfund Insider, interviewing Miyachi in May 2025, described a founder based in Austin who left Amazon to build systems that respond faster to threats such as synthetic media.
The product is BitMind Forensics. In an arXiv paper of 14 July 2026, written with Dylan Uys, the system is trained through an open adversarial competition with three roles: generative miners supply synthetic media, validators check it, and discriminative miners submit detector checkpoints that compete to classify it. New generators enter the challenge within hours, and the best detector in each modality becomes the basis of the production export. On the Deepfake-Eval-2024 benchmark the authors report an area under the curve of 0.915 for images and 0.822 for video. They also state limits: weaker results on older, compression-heavy face-swap datasets, and audio deepfakes left out of scope.
The argument behind it recurs in Miyachi's public writing. In an opinion piece for crypto.news on 20 May 2025, Miyachi wrote that "decentralization rewires AI's incentives so that rewards and governance flow to the true value creators rather than bottling up inside a single balance sheet." Interviewed by Crowdfund Insider, Miyachi put the detection problem this way: "The fundamental problem is that conventional detectors are static while generative AI is dynamic."
Miyachi spoke at Proof of Talk Paris 2026, held on 2–3 June at the Musée des Arts Décoratifs, Louvre Palace. On 2 June the programme placed Miyachi in the Bittensor Track subnet keynotes on the Taostats Stage, 15:30 to 16:45, alongside Robert Myers, Julien du Bois, Micaela Bazo, Tim Kalic and Shardul Bansal. On 3 June came TensorFi: Unlocking DeFi for the Bittensor Economy, a side event on the X Ventures Masterclass Stage, 12:30 to 13:15, with Mike Grantis.
What Ken believes about the next ten years.
Years of shipping.
Lines that keep proving correct.
“The fundamental problem is that conventional detectors are static while generative AI is dynamic.”
“In short, decentralization rewires AI's incentives so that rewards and governance flow to the true value creators rather than bottling up inside a single balance sheet.”
What they have built, written, and backed.
Built
- BitMind ForensicsA deepfake detector made up of three checkpoints, an image ensemble, a general video model and a face-cropped human-video model, trained through Bittensor Subnet 34. An arXiv paper of 14 July 2026 reports 0.915 AUC on Deepfake-Eval-2024 images and 0.822 on video.
- Bittensor Subnet 34 adversarial competitionAn open competition in which generative miners supply synthetic media, validators check it and discriminative miners submit detectors, with the best detector in each modality adopted as the basis of the production system.
Ken at Proof of Talk Paris 2026
The next Proof of Talk is in Abu Dhabi.
Louvre Abu Dhabi, 3–4 December 2026. Admission is by application, and the room is capped at 2,000.
Request access