While business leaders are chasing productivity boosts and cost reductions by investing heavily in AI, new evidence suggests that many are sleepwalking toward significant governance failures, according to new BSI research.
The global study – combining an AI-assisted analysis of over 100 annual reports from multinationals and two global polls of more than 850 senior business leaders, conducted six months apart – offers a comprehensive view of how AI is publicly framed in corporate communications, alongside executive-level insights into its implementation.
According to the study, 62% of business leaders expect to increase investment in AI in the next year, and when asked why, a majority cited boosting productivity and efficiency (61%), with half (49%) focused on reducing costs. A majority (59%) now consider AI to be crucial to their organization’s growth, highlighting the integral role executives see AI playing in the future success of their businesses.
Highlighting the striking absence of safeguards, fewer than a quarter (24%) reported that their organization has an AI governance programme, although this rose modestly to just over a third (34%) in large enterprises – a pattern repeated across the research. While nearly half (47%) say AI use is controlled by formal processes (up from 15% in February 2025), only a third (34%) report using voluntary codes of practice (up from 19%).
Additionally, just a quarter (24%) say employee use of AI tools is monitored, 30% have processes to assess the risks introduced by AI and the required mitigations, and only one in five businesses (22%) restrict employees from using unauthorised AI.
A key component of AI governance and management lies in how data is collected, stored, and used to train large language models (LLMs). Yet only 28% of business leaders know what sources of data their organization uses to train or deploy its AI tools, down from 35% in February. Just two-fifths (40%) said their business has clear processes in place around the use of confidential data for AI training.
