Two brain networks were measured here, in largely the same participants and on the same set of yardsticks. Largely, because the executive task is missing for two people, one older cohort did a slightly different version of it, and the connectivity measures rest on subsets. Five measures were applied to both. The language network showed no detectable difference on three of them, and differed on the other two, in one case in a direction the authors read as capability rather than decline. The executive network differed from the younger comparison group on all five.
The study appeared in Nature Communications on 24 August 2026, from a group spanning MIT, Boston University, Harvard, Massachusetts General Hospital, Stanford, UCLA and UC Merced. It compared 64 older adults, average age 59.7 and ranging from 41 to 80, against 483 younger adults averaging 23.8 and ranging from 17 to 39. All of them were native English speakers recruited around Boston, with normal or corrected vision and hearing, and none reported any psychiatric or neurological disorder. That is a screened, healthy, self-selected research sample, not a cross-section of people growing older.
Why these two systems keep getting confused
The language network is a set of left lateralised regions in frontal and temporal cortex that respond to meaningful, structured language and not to much else, though the authors note it necessarily interacts with other brain systems during real-life language use. The Multiple Demand network is domain general: it engages whenever a task gets hard, whatever the task is. In the left frontal lobe the two lie close enough together to be confused.
The authors argue that this is part of why the ageing literature disagrees with itself. They name several contributing factors; two of them matter here. One is task conflation. Many language paradigms recruit both networks at once, so an older adult who finds the sentences effortful will engage more executive cortex, and the resulting map looks like a language network that has spread out with age.
The other is averaging. Standard fMRI warps everyone onto a template brain before averaging, and because these regions sit in slightly different places in different people, that approach “blurs network boundaries, leading to interpretive challenges.”
Precision fMRI takes the other route. Each participant first runs an extensively validated localiser task, and their own peak responses define their own regions. Here the language localiser was passive silent reading of sentences against lists of pronounceable nonwords. The executive localiser was a spatial working memory task, holding eight locations in a three by four grid against an easier version holding four.
Within each search space, the top 10 percent most responsive voxels in that individual became that individual’s regions. Response magnitudes were then read out from a different run of data than the one used to define them. Executive localiser data were missing for one older and one younger participant.
The five measures, side by side
The first was how closely each person’s map matched a normative atlas built from hundreds of other people. Language: no group difference. Executive: significantly less typical in the older group.
The second was how far each person’s peak responses sat from the atlas peaks. Language: peak locations were, on average, slightly less scattered in the older group, not more, and for the language network this was measured on left hemisphere parcels only. Executive: significantly more scattered.
The third was how much cortex responded at all. Language: no difference. Executive: significantly less, both inside the network’s search spaces and across the whole brain.
The fourth was how strongly it responded. Language: the older group showed a significantly higher response to the sentences over nonwords contrast, an effect the paper says appears to be mostly restricted to the left posterior temporal region. The size of it is small; the paper puts the average effect in younger adults at 90.5 percent of the effect in older adults. Executive: significantly weaker, and the reduction was driven by the hard condition, with no group difference on the easy one.
The fifth was how tightly regions within a network moved together over time, and it rests on subsets rather than the whole sample: 53 of the older adults at rest and 38 during story listening, against 82 of the younger adults in both. Language: within the left hemisphere network, no difference detected, at rest or during story listening. Executive: significantly weaker in both.
A sixth measure was applied to the language network only, because, as the paper puts it, the executive network does not show a strong hemispheric bias. Left hemisphere dominance: no difference detected.
One thing to hold onto about all of those nulls. The paper’s main text reports no equivalence test and no Bayes factor. So “no difference” throughout means no difference was detected between 64 older adults and 483 younger ones, which is not the same claim as no difference existing.
Thirty eight of the older adults also listened to a story of about five minutes while being scanned. Their language regions tracked word frequency and word to word predictability, mirroring what the authors say has previously been reported for young adults. Afterwards they answered seven yes or no comprehension questions with an average accuracy of 97 percent, standard deviation 10.3.
Three differences the announcement leaves out
The popular version of this result, including the one in the university’s own announcement, is that no differences were found in the language network. Anne Billot, described there as one of the lead authors and in the paper as one of two who contributed equally, is quoted saying: “In the language network, we couldn’t find any differences between older and younger groups.”
The paper reports at least three within the language network proper. Response magnitude is higher in the older group, and the paper itself calls this “the only small but significant difference we observed between age groups within the language network.” Peak scatter is slightly lower. And connectivity between the two hemispheres is higher in the older group at rest, though not as a function of continuous age.
Three more sit just outside that boundary. Cross-talk between the language and executive networks rose slightly at rest in the older group. And in the wider band of language-responsive cortex, the older group responded less in one region, the bilateral medial anterior superior frontal gyrus, and more in a part of the cerebellum; the authors did not predict either and call them preliminary. The lower medial frontal response matters for how the raised response elsewhere gets read: this paper takes a higher language response as reflecting older adults’ larger vocabulary and greater proficiency, while other work has read raised responses in language regions as a sign of effort. The paper cites studies reporting age-related increases, but does not weigh an effort account against its own. It is a reading, not a measurement, and the study does not settle which is right. What it does report is that the two networks stay robustly segregated in older adults, which it offers as evidence against the idea that ageing blurs specialised systems together.
The same announcement says the researchers identified the language network by having participants “listen to stories and read sentences.” The localiser was silent reading alone. Story listening was a separate task, given to 38 of the 64 older adults and to 82 of the younger ones, and the two groups did not even hear the same story.
There is a wider gap between the popular framing and what was tested. This study measured brain responses during comprehension, and only passive comprehension. The everyday complaint people actually have about ageing and language is word finding, and word finding is production. The paper is clear that production is a real exception, noting that “some aspects of language production (specifically, word retrieval) exhibit age-related decline.” It is equally clear that its own design does not reach it: whether the similar response magnitudes “would generalize to production remains to be established.”
It offers a hypothesis about that gap rather than an answer. The weakening executive network, the authors suggest, “may produce downstream effects on language use even when the language network itself is intact,” under time pressure, in dual task settings, or during effortful retrieval. That is a proposal for future work.
Three further limits belong beside the headline. The design is cross sectional, so nobody was followed over time and nothing here tracks a single person’s network as that person ages; the authors note they may be “missing complex non-linear changes across the lifespan.” Cognitive screening was carried out in only one of the two older cohorts, 38 people of 64. And education data were not available for this study at all, with the authors listing education, occupation complexity and generational experience among the things a future study should match samples on.
Two smaller wrinkles are worth knowing. The younger adults and one older cohort were scanned on one machine and the second older cohort on another after an upgrade, which the paper addresses by reporting that standardised preprocessing makes the two produce near identical results. And the core finding is not brand new: a preprint has been public on bioRxiv since October 2024, under a title beginning “The language network ages well.” The headline has therefore been in the open for nearly two years.
Ageing, in the plural
The habit this study interrupts is the singular one. There is a widespread way of talking about the ageing brain in which decline is a property of the organ, spreading outward at some rate, with every function downstream of it.
What two networks measured on the same yardsticks in the same skulls suggest instead is that age does not meet every network on the same terms. The paper’s own conclusion is that the field needs “more nuanced models of brain aging that consider the distinct trajectories of different brain networks.” A system that spends a lifetime accumulating knowledge and a system that supplies flexible effort on demand appear to meet age on quite different terms.