The OECD sets a common AI baseline ahead of binding rules

Forty-two countries adopted OECD principles for trustworthy AI and the G20 followed. Explainability and accountability are becoming due-diligence items.
Forty-two governments now share one definition of trustworthy AI. On 22 May the 36 member countries of the OECD and six others, Argentina, Brazil, Colombia, Costa Rica, Peru and Romania, adopted the OECD Recommendation on AI, the first intergovernmental standard on the technology. On 9 June the G20 trade and digital economy ministers, meeting in Tsukuba, Japan, adopted non-binding G20 AI Principles drawn from the OECD text.
The principles bind no one, but they set the vocabulary that binding rules will use, and that moves explainability and accountability from ethics statements into due diligence. Governments in the United States, the European Union and Latin America have signed the same text, and the G20 version extends it to a group that includes China and India. Each element becomes a question an acquirer, a bank or a public buyer can ask: can the company explain a decision, show how it tested the system and name who is accountable when it fails?
The text is concrete enough to be used that way. Its five principles cover inclusive growth and well-being, respect for human rights and democratic values, transparency and explainability, robustness, security and safety, and the accountability of those who build and operate AI systems. Five recommendations to governments cover public investment in research, a digital ecosystem of data and infrastructure, an enabling policy environment, skills and international cooperation. Fanaura expects the price effect to arrive long before any law takes effect: a company that can document these qualities sells faster to regulated buyers and earns a better price at exit, while one that cannot carries a discount.
The cheapest moment to build for these principles is before the product scales. That means records of training data and its origin, tests for bias and failure before release, logs of the decisions systems make and a named owner for every model in production. Founders selling into finance, health, hiring or the public sector should expect these questions in procurement soon. Because five Latin American countries signed alongside the OECD members, the same documentation travels across the markets where Fanaura's mandate applies.
The obvious objection is that a non-binding text changes nothing until someone enforces it. The coming months will show how much weight it carries. G20 leaders can endorse the principles at their Osaka summit at the end of June. The European Commission has to decide how to follow up the ethics guidelines for trustworthy AI that its expert group published in April, and whether the next step carries obligations. And the way the OECD tracks implementation across the 42 adherents will separate governments that treat the text as policy from those that treat it as a statement.
AI regulation is starting from shared principles rather than from separate national rulebooks. That favors companies that build explainability, testing and accountability into the product early, since the cost of retrofitting rises with every customer, a premise of our venture thesis.
Building an AI-native company? Apply in five minutes. →
More from Fanaura
Building AI systems that can explain and document their decisions? Send us the project.
Fanaura invests its own capital and the capital of its limited partners (LPs), usually through a separate SPV for each deal. If we see a fit, we get in touch.
Get the news by email
One short note each weekday. No marketing, and you can unsubscribe at any time.