The two percentages that will end up in a governance deck measure different objects, and the difference decides what a board actually has to do about it. Glass Lewis counts clear attribution of cybersecurity oversight, meaning a disclosed statement of which body owns the topic, at over nine in ten large caps in the UK and Continental Europe [1], and defined board oversight of AI at around seven in ten European large caps [2]. The American figure in the same memorandum, roughly 21% of Russell 1000 companies with meetings through June 2026, counts something narrower, which is having an AI policy in place [6]. A company can name an owner without writing a policy, or write a policy without naming an owner, and only the first appears as a line in a committee charter.
The movement is where the two regions separate. Continental European large caps went from around one in five with an AI policy in 2025 to more than half [3], UK companies reached over four in ten [4], and the Russell 1000 moved from about 15% to 21% [6]. That is roughly 30 points of year-over-year change against six, about five times as much [1]. Inside Europe the catch-up is still unfinished, with about 20 points separating cyber attribution from AI oversight [2].
A skeptic on the audit committee would say none of this has cost a director a seat, and on the 2026 record that is correct. Sixty-three US director nominees failed to win majority support, down from 72 in 2025 [7], and only four of them left their boards while 11 resignations were rejected, a departure rate near 6% with 59 directors still seated [8][3]. Canada recorded two failures, down from ten [9]. Australian dissent reached a ten-year high, with 26 ASX300 directors drawing over 25% opposition, nine more than in 2024 [10][4], and Glass Lewis attributes that to share performance, operational challenges, capital allocation and concerns over board composition [11]. In Europe the share of companies with a director proposal above 20% opposition rose from about 2% to about 5% while no uncontested large-cap election failed [12]. The vote channel this season registered performance, not disclosure practice.
The board-deck version says an AI working group exists and will be formalised later in the year, and it is incomplete in one specific respect: the comparison has moved outside the company. The counting appears in a Glass Lewis memorandum covering multiple markets [13], and the peer set has already answered the question in writing. Naming an owner costs agenda time and creates a duty the board then has to evidence at each meeting, while leaving it unnamed costs nothing this quarter and turns into a question a voter can ask with a peer number attached next year. The decision available now is which existing committee absorbs the topic, because the next cycle's comparison runs against a European baseline near seven in ten rather than against an empty field [2].
