An office administrator who reads the January report to the end finds their own occupation scored between 60% and 68% for AI exposure, up from a range of 14% to 21% three years earlier [8]. There is no paragraph explaining why that score will not fully land.
The chief executive row has one. Cognizant had previously forecast CEOs reaching 25% exposure by 2032; the update puts them above 60%, covering contract negotiation, policy recommendations and implementing those policies [12]. The report then stops to say that whether boards and shareholders would accept those tasks being automated is another question, and that accountability will likely constrain how far the theoretical exposure plays out [13]. The argument is sound, and it is also, as TNW points out, the only place a named group gets a real-world brake fitted to its score: nothing comparable sits beside the financial managers at 84% or the admin floor [14][9].
The construction of the headline figure explains why the brake matters. Cognizant took Bureau of Labor Statistics headcounts, multiplied them by median salaries, and applied its exposure scores on top [4]. That arithmetic captures what a job pays and what a model can attempt, but not who is answerable when the output is wrong, and that is exactly the variable the CEO paragraph introduces and then applies to one row only.
The hiring side is the same figure with headcount attached. The graduate intake is 10% of the 15,000 the Frontier job family is meant to reach, and the announcement as reported does not say where the other 13,500 come from [2]. Nor does the material show how those roles get priced or billed to clients, which is the part an operator buying delivery would most want to see. What it does show is the asymmetry in who gets named: the incoming graduates get a titled category built for them, while the occupations already scored at 60% and above get a percentage.
For anyone handed an exposure percentage next quarter, the CEO paragraph is the template to copy: the specific mechanism that would stop the automation, whether that is who signs, who carries the liability, which licence, which audit trail. Where such a mechanism exists, it is the brake the report gave chief executives, and the score becomes a ceiling that is never approached. Where it does not exist, the role is either genuinely automatable or simply has nobody senior enough to write its caveat, and those two situations call for different responses. Lawyers moved from 9% to 63% in three years [10], a sevenfold rise [3]. That is a finding about legal work, and about how fast a capability score travels when no one defends the row.
