Doreen Bogdan-Martin, 2023

“We need guardrails to ensure AI is safe, responsible and inclusive.”

International Telecommunication Union Secretary-General Doreen Bogdan-Martin called for these guardrails at the 2023 AI for Good Global Summit. The phrase is short, but its three adjectives carry separate demands. Safety concerns whether a system behaves reliably and resists misuse. Responsibility concerns who is accountable for its design, deployment, and consequences. Inclusion concerns who can access its benefits, whose languages and needs are represented, and who participates in decisions about it. Treating any one of the three as optional can produce a system that succeeds technically while failing the public it is meant to serve.

The ITU’s role makes inclusion especially significant. Billions of people still face gaps in connectivity, affordability, digital skills, and access to modern infrastructure. An AI service cannot improve agriculture, education, health, or disaster response for a community that lacks dependable networks or local-language resources. At the same time, expanding access without protection can expose new users to fraud, surveillance, manipulation, or unsafe products. Digital development and AI governance therefore have to advance together. The objective is not merely more connections, but connections people can use with confidence and agency.

Guardrails also need technical substance. They can include predeployment testing, security controls, documentation, incident reporting, privacy protections, and monitoring under real conditions. They may also include limits on applications that cannot yet meet a reasonable safety threshold. International standards help different countries and vendors compare evidence, but standards are not self-enforcing. Procurement rules, regulators, independent researchers, and civil society all have roles in checking whether a claim made in a policy document remains true after a system reaches users.

Bogdan-Martin’s line is a reminder that global AI policy cannot be reduced to a competition among the largest companies and states. Smaller countries, local researchers, women, young people, people with disabilities, and communities speaking underrepresented languages bring knowledge that centralized development often misses. Inclusion improves both fairness and performance because it exposes assumptions before they become infrastructure. Responsible AI is not achieved by inviting diverse users after a product is finished. It begins by sharing the power to define the problem, the acceptable tradeoffs, and the evidence required for deployment. A guardrail is valuable because it allows movement while marking a boundary. The best AI governance does the same: it supports useful experimentation, states where evidence is insufficient, and prevents speed or market pressure from pushing people beyond protections they cannot replace themselves. Its strength is proven under pressure.