Tertium AI

Tertium AI

Principles

Tertium’s principles are defined as either entrenched or operating. The entrenched principles are locked by our governance plan: Tertium cannot act against them, and any addition, change or removal requires the consent of the independent Tertium board. The operating principles reflect our current thinking; we expect to revise them as the evidence and the technology develop, and we will publish every change and the reasoning behind it.

Entrenched Principles

  • Tertium is a non-profit.
  • Tertium exists to build and distribute frontier AI. Providing inference will be left to downstream consumers of the model weights.
  • We’re open by default: open code; open data, subject to privacy and copyright; and open weights wherever responsible. We release as widely as we can demonstrate to be safe.
  • Where we can’t release weights openly, we share them with our partners: democratically aligned nations, enterprises or individuals that share our values and help carry the cost of building the models.
  • Access for partners is governed by just two things: being a contributing member of the programme, through capital, data, compute or expertise; and having the required cyber-defensive capability to limit the risk of weights leaking.
  • The terms are the same for all partners, and access is never used as a lever. Membership is binary: paying in more than your share doesn’t buy earlier access, more models, or any other advantage.

Operating Principles

  • AI is set to be the most consequential technology humanity has ever developed. Our purpose is to put sovereign access within reach of as many countries and their peoples as possible, giving them the agency to shape their own future.
  • Members contribute to the cost of building, but no one has a monopoly on the shared capability; our aim is to narrow the gap between nations.
  • Narrow data builds narrow models. We train on the breadth of human knowledge: the world’s languages, cultures and ways of thinking so that the future they shape is inclusive.
  • Balancing safety and capability is one of the most important judgements Tertium has to make when targeting transformational capability. Powerful models deployed insecurely, or developed too hastily, could significantly increase the risks AI poses to humanity.
  • We build systems that remain under meaningful human oversight. Our models are designed to be governable and to answer to human judgement.
  • Before any model is considered for release, our Safety Board commissions a safety report from external evaluators that are independent of any single member, geographically diverse by design and expert in eliciting dangerous capabilities. Bodies doing this work today include the UK AISI, METR and Apollo Research. The same board commissions the independent verification of members’ cyber-defensive capability.
  • Tertium owns the release decision: whether a model goes out open-weight or shared-weight, and if shared, what cyber-defensive standard members must meet to receive it. We commit to explaining why we reached each decision, given the board’s report.
  • Release decisions aren’t static. We move models from shared-weight to open-weight as the external open-weight frontier advances, and from higher to lower security requirements as the marginal risk of their leaking falls.
  • To release models as openly as possible, we work to reduce the chance of misuse by malicious actors, building in interventions such as bio-risk data filtering and greater resistance to jailbreak and fine-tuning attacks. Even so, our most capable models will likely sit in the shared-weight tier.
  • We will keep a clear separation between our decision-making and our funding. The Tertium board and the Safety Board are funded independently of the executive, and changes to that funding mechanism require the Tertium board’s consent. We will retain autonomy over balancing safety against capability, and will maintain the ability to slow development when required.
  • We govern our organization by a plan published in advance. The plan gives a transparent scaffolding around our org structure and decision-making so that we can be held accountable for our own guidelines, as a commitment is only worth as much as it is verifiable and costly to break.
  • We’re uncertain about the moral status of future AI systems, and commit to investigating it as a serious open question.

Read the overview

How Tertium is structured, how partner access works, and how we decide whether a model goes out open-weight or shared-weight.

Tertium Overview →