AI Openness Report: 3 years on from Llama 2
- China releases frontier-level open weight Kimi K3 model released on third Anniversary of Llama 2 opening up sees open model taking on proprietary frontier model
- US policy shifts as a result Sovereignty and rise of China’s structured open source strategy
- UK has pipped Germany into 3rd place on AI Repo growth
- Agentic Harness and Infrastructure being built by individuals and innovators on GitHub
- Huge growth in agentic projects – of the 891 projects with 100+ stars, 40% of agentic repositories had none 12 months ago
London, UK – 23 July 2026 – OpenUK, the non-profit organisation convening the UK’s Open Technology sector, and its international sister OpenHQ, announce their latest AI Openness report, a retrospective of 3 years of AI openness, summarising the technological, policy and value shifts as open source went from a technical choice to a strategic option for enterprises and nation states navigating a sovereignty-focused world. The report navigates the what, the how, and the why around AI and open source, including open source becoming a lever of influence and soft power. The report also goes into why the UK should have its own Foundation.
Findings:
- There has been a significant shift in expected value and return on investment – companies are looking at customising models to suit them rather than using frontier AI models to achieve their goals
- Even closed models want open standards and infrastructure – this has led to the new foundations like OpenClaw Foundation and Agentic AI Foundation holding these projects o behalf of their communities, playing a critical role in how to support AI adoption and empower their owners and leaders
- Soft power around AI continues to evolve – while the US and China have the leading companies in their countries, other countries and regions are looking at how to strengthen their own positions, affecting both current and future sovereignty approaches
- There is a call for an expert-led UK Foundation around open source and AI projects, to build on UK’s demonstrable innovation and follow the Chinese model of planned growth around open source
The data shows
- The UK has surpassed Germany as Europe’s leader by number of AI repositories in open source
- Models are being shared by companies as open weights in a less collaborative model
- The Agentic AI harness projects for models are being built by individuals and communities in a more traditional open source collaboration – this means they tend to be on GitHub where collaboration is easier compared to HuggingFace, which is more for sharing models
- The UK has spawned some of market leaders in open source agentic AI
- Model value is in contention with the biggest models no longer seen to be best – customisable models matter more
- Dev Tools and infrastructure are seen as critical areas for AI monetisation and return on investment
Case Studies
- Open weights, owned compute and the case for sovereign AI, James Drayson, CEO and Co-founder, Locai Labs
- Agentmemory, Rohit Ghumare, Creator & Maintainer, Agentmemory
- Open weights, owned compute and the case for sovereign AI, Tom Wilkinson, Chief Data Officer, Scottish Government
- The OpenClaw Foundation, Vincent Koc, Chief Architect, OpenClaw Foundation
Thought Leadership
- The Commons we have not (yet) created, Anastasia Stasenko, Founder and CEO, Pleias
- Open Source tools for AI, Hannah Foxwell, Co-founder BIMP and creator of AI for the rest of us
- Security and Vendor response, Karan Saini, Research Manager, OpenUK
- The Entrepreneur’s AI Journey, Matt Barker, Founder and CEO, Bolt MCP
- The EU Open Source Policy, Sachiko Muto, Chairwoman, Open Forum Europe
- Security, Identity and Harnessing Agentic, Sal Kimmich, Gadfly AI
And Fireside Chats
- Open foundation models enabling a map of the planet, Professor Anil Madhavapeddy, Professor of Planetary Computing, University of Cambridge
- Mobile Context Protocol (MCP) and Protecting its openness, David Soria Parra, Member of Technical Staff, Anthropic
Amanda Brock, CEO at OpenUK, commented:
“The data tells us that the UK is not well served by throwing money at building new large or frontier models. The opportunity to hone in on with a laser focus is the model-adjacent technologies in the AI harnesses, infrastructure and developer tooling. The lower barrier to entry is enabling individuals and innovators to break into this space.
Collaborating through their traditional GitHub platform over Hugging Face they are building the adjacent technologies that are becoming the open standards of AI and even the proprietary model owners want a piece of that action, setting up US foundations to host them and building soft power.
Perhaps ‘selling picks and shovels’ of the “AI gold rush” isn’t as sexy and it definitely isn’t as leveraged, but may well be where we see the ROI. For the UK we are leaders in this deep tech and it’s more suited to our short pockets. Rather than competing with the deep pockets of the US and China, this is where middle nations have an opportunity and the UK in particular is well placed. .
We need to mimic the soft power in this space that is rapidly establishing in the US with our own Foundation for open source projects. That’s not a lab or a policy shop but a technical base built by our ecosystem’s deep experts rather than continuing to build the pipeline of innovation for other nations to leverage.”
Supporting Quotes:
James Drayson, Co-Founder and CEO at Locai Labs comments in the report that, “For organisations already worried about control, data security, intellectual property leakage and unpredictable API pricing, [these] policy positions have converted a hypothetical concern into a reality and a precedent. This has prompted a movement back towards on-prem deployment in the UK and other worried nation states. Openness is precisely what turns a capable model into a trusted foundation. The ask is simple: if you want enterprises to build on your model, show them what it’s made of.”
David Soria Parra, Member of Technical Staff, Anthropic and one of the creators of MCP, commented: “Protocols are definitions for how systems communicate with each other. They only work if they are open and can be freely implemented. The alternative is to do a proprietary API. You can do that, but we believe that an open source protocol like MCP (Model Context Protocol) uplifts everyone in the ecosystem, and allows people to use AI more as a whole … Open standards keep AI companies competing on model quality and safety, not on who controls the tools. Open source software is essential for building a secure and innovative ecosystem for agentic AI.”
Vincent Koc, Chief Architect, OpenClaw Foundation, commented in his submission for the report: “Open agents help democratise access to AI, and access here includes the proprietary and enterprise models too. Because OpenClaw gives everyone the same open starting point, switching between models and tools becomes easy … The bigger takeaway is what access does – democratising access to AI, making it available to everyone in a form they can actually use and own, strengthens the economy, education, and innovation across every sector of a country. OpenClaw is proof that this can happen bottom-up, from a community, without anyone’s permission.”
About OpenUK and OpenHQ
OpenUK is the organisation for the business of Open Technology, being Open Source Software, Open Source Hardware, Open Data, Open Standards and AI Openness across the UK. Its purpose is UK leadership and global collaboration in Open Technology. OpenUK works on three pillars: Community, Legal and Policy and Learning. In 2025 OpenUK focused on AI and openness and soft launched its sister organisation OpenHQ to focus on coalescing international communities such as that in India for international policy initiatives.
OpenUK is a not-for-profit company limited by guarantee, company number 11209475.
Twitter: @openuk_uk
Bluesky: @openuk.bsky.social
Mastodon: https://hachyderm.io/@openuk
LinkedIn: openukopentechnology
Website: openuk.uk and theopenhq.com/
