impacting wages and hiring. About 1 in 10 job vacancies in advanced economies demands at least one new

skill, often appearing first in the United States. The incidence is about half of that in emerging market

economies. These skills boost average wages and employment but deepen polarization, mostly benefitting

high- and—through higher consumption of services—low-skilled workers, and potentially contributing to the

shrinking of the middle class. Vacancies demanding AI skills post higher wages, but the diffusion of such skills

is linked to lower employment in occupations with high exposure and low complementarity to AI, posing

challenges for the youth. A Skill Imbalance Index reveals wide cross-country differences. Economies facing

strong demand should prioritize education and reskilling, while those facing strong supply should foster firms’

absorption through innovation and access to credit.

and understanding the motivations of modern reasoning models. However, if

models can control what they verbalize in their CoT, it could undermine CoT

monitorability. To measure this undesirable capability — CoT controllability —

we introduce the CoT-Control evaluation suite, which includes tasks that require

models to solve problems while adhering to CoT instructions, e.g., reasoning about

a genetics question without using the word “chromosome”. We show that reasoning

models possess significantly lower CoT controllability than output controllability;

for instance, Claude Sonnet 4.5 can control its CoT only 2.7% of the time but 61.9%

when controlling its final output. We also find that CoT controllability is higher

for larger models and decreases with more RL training, test-time compute, and

increased problem difficulty. CoT controllability failures extend even to situations

in which models are given incentives (as opposed to direct requests) to evade CoT

monitors, although models exhibit slightly higher controllability when they are

told they are being monitored. Similarly, eliciting controllability by adversarially

optimizing prompts does not meaningfully increase controllability. Our results

leave us cautiously optimistic that CoT controllability is currently unlikely to

be a failure mode of CoT monitorability. However, the mechanism behind low

controllability is not well understood. Given its importance for maintaining CoT

monitorability, we recommend that frontier labs track CoT controllability in future

models.