3.3 Responsible AI AI technology is evolving rapidly, and while regulators play a critical role, CSPs must self- govern using their own responsible AI principles and taxonomy until broader regulations and use cases mature. The World Economic Forum telecommunications community highlights the imperative of maintaining regulatory compliance, trust credentials and service reliability. Where AI implementation impacts their ability to provide critical communication services, they face the risk of severe penalties and reputational damage. AI-related risks span business functions and include intellectual property (IP) exposure, data bias in automated decision-making, inadequate outcomes (for example, in automated customer service responses) and confidentiality or privacy risks. To manage these risks, a clear understanding and control over the data foundations that drive AI predictions and automation are essential. Region-specific standards and expectations also add complexity, with varying definitions of “sensitive data”. The use of LLMs introduces further challenges, such as ensuring social inclusivity through multilingual capabilities, accessibility and sustainability. Sustainability is particularly pressing. Public corporate social responsibility reports show that 68% of the global industry (by revenue) must significantly cut emissions by 2030 to meet targets.43 Incremental energy use from LLMs should be viewed through this lens, balanced by energy-saving use cases and SLMs to reduce both energy and compute costs. To aid self-governance, industry bodies have provided resources such as Amazon Web Services (AWS) and TM Forum’s Generative AI Maturity Interactive Tool44 and The GSMA Responsible AI Maturity Roadmap.45 These tools help CSPs identify and address current process gaps. The World Economic Forum’s Digital Trust Framework can also be used to consider the security, reliability, accountability, oversight and ethical use of digital technologies.46 Artificial Intelligence in Telecommunications 21

Artificial Intelligence in Telecommunications 2025