Gensyn launches open-1b, an AI model built to be independently audited

Gensyn said it launched open-1b, a verifiable AI model meant to let third parties independently check how the model was trained. The London company frames the release as “auditable AI” — not just open weights or an open training recipe, but proof that others can re-verify training steps across different hardware.

Same-day SAFETY primary that a lab shipped a model package claiming third parties can independently verify training — pushing transparency past open weights into auditable proof, if the company’s method holds up.

Gensyn, a London AI lab, announced the launch of open-1b via PR Newswire on 15 Sep 2026. The release is datelined LONDON, Sept. 15, 2026, and timed 08:08 ET. The company calls open-1b a verifiable AI model designed so anyone can test and independently check how it was trained. That company PRIMARY via PR Newswire is the filing event. These are company claims. This desk did not train or audit the model.

Positioning, as Gensyn tells it: closed models such as Claude and ChatGPT give outputs only, and keep the training method proprietary. Open-weight models such as Llama share the weights — the downloadable numbers that make the model run — but not the training process. Open-recipe models such as OLMo share the weights plus the instructions used for training. Auditable AI, the company says, is what open-1b is meant to be: the model, the training recipe, and verifiable proof so third parties can independently audit any stage of training. File that four-rung ladder as Gensyn’s. The company’s “industry’s first auditable AI model” and “industry’s leading decentralized AI lab” lines stay company claims. This desk is not independently certifying that no other auditable model exists.

Problem framing, still company: simply sharing code and datasets is not enough for true independent verification, because hardware variation often produces inconsistent results. Gensyn says its infrastructure makes training outcomes reproducible and verifiable across diverse hardware, and that open-1b demonstrates that capability. File the hardware-variation problem and that reproducibility claim as Gensyn’s. This desk did not rerun training on a second chip.

Named voice on the release: Ben Fielding, co-founder and chief executive. He argues auditable AI means a model’s decisions do not have to come from a black box, and that someone should be able to inspect the data that shaped the model, verify the training process step by step, and understand why it reached a particular conclusion. File the name and title as the company’s. His quotes are color only and stay in Sources.

Backers named on the same wire, company: a16z crypto, CoinFund, Galaxy Digital, Eden Block, Maven 11, and more. File those names as Gensyn’s. This is a model launch, not a funding-round announce. This desk is not inventing a raise size, a valuation, or a close date. None was printed.

Company context, still Gensyn’s: the firm describes itself as a decentralized AI lab building forecasting systems that improve by learning from real-world outcomes, pairing verifiable infrastructure with incentives for people to contribute information models have not seen. File that mission box as the company’s. This filing is the open-1b launch only.

Plain English for the rest of the card: open weights = downloadable model parameters, the numbers that make a model run. Open recipe = published training instructions and data recipe. Auditable, as the company uses it = third parties can check and prove how training ran, not only download the finished weights. 1b in the name is the company’s about-one-billion-parameters class label — not a benchmark score this desk measured.

PRIMARY here: Gensyn’s 15 Sep 2026 PR Newswire company release — Tier A company source, the original record. The company site is product-home context, not a second originating newsroom. The open-1b launch, the closed / open-weight / open-recipe / auditable ladder, the hardware-variation reproducibility claim, the Fielding name, the named backers, and the forecasting-lab mission box are company-attributed. The company’s “first,” “industry-leading,” and “new standard for transparency” lines stay company-attributed. NOT claimed: that this desk independently certified open-1b as the first auditable model, independently verified training proofs, benchmark scores, download numbers, a raise size, a valuation, that this desk retrained the model, a stock tip, or investment advice. Distinct from the already-filed nofire-brig-open-source and china-tc260-ai-safety-framework-3-0.