Peter Allen Mann

Signs of Autism in Women

How a high-masking profile gets missed for years

The criteria were written down in 1943 and 1944, from two clinic caseloads, and almost every child in them was a boy. Everything downstream of that is a filter. The filter is still running.

I was fifty-five when anyone pointed it at me, and I am a man. So this essay is not written from the inside of the most-missed version of the profile. It is written from thirty-seven years of reading performance files, hiring, promoting and losing people, and from the four decades before my own recognition when the filter was running on me and no one looked.

What this essay is and isn’t. I’m not a clinician. I’m an operator with a diagnosis. Nothing here is a screening tool, a checklist that resolves anything, or a route to an answer about yourself. Anyone who wants that answer needs a clinician and a proper assessment. The research below is published and linked. The rest is my take.

Key takeaways

  • The criteria that decide who gets assessed were built from a narrow sample, and a presentation that falls outside that sample reads as ordinary rather than as autistic.
  • Camouflaging is the mechanism, and it is measurable. It is also socially rewarded, which is why it strengthens over decades instead of lapsing.
  • Competence is the most effective way to stop anyone asking what something costs. Delivery is the only signal most workplaces collect.
  • The research on this presentation is now good enough to quantify the mental-health cost and still has a named hole in it, and the hole sits exactly where the late-recognized population lives.
  • Recognition is not a treatment. What it changes is which question you ask about a bad week.

Contents

What the criteria were built from

The diagnostic description of autism began as a description of a particular group of children seen in particular clinics. The samples were small and overwhelmingly male, and the behaviors recorded in them became the reference set. Decades of refinement later, the criteria are far better, and they are still anchored to what that reference set made visible.

This is an ordinary measurement problem, and it is the kind I spent a career on. An instrument built from one population measures that population well. Run it on a second population whose presentation differs and it does not return a different answer. It returns no answer at all, which reads as a negative.

The clearest published evidence that this is happening is the ratio itself. Loomes, Hull and Mandy pooled fifty-four studies and found the true male-to-female ratio of autism closer to three to one than the four to one that clinical records report, and their conclusion was blunt: girls who meet criteria are at disproportionate risk of not being diagnosed (Journal of the American Academy of Child & Adolescent Psychiatry, What Is the Male-to-Female Ratio in Autism Spectrum Disorder?). The gap between the measured ratio and the recorded ratio is the size of the filter.

So the honest framing is not that there is a female autism. It is that there is a presentation the instrument was not built to see, it appears more often in women and girls for reasons that include how differently they are socialized, and it also appears in plenty of men. I am one of them. Writing it as a category belonging to one sex makes the mechanism harder to see, and the mechanism is the useful part.

What are the signs of autism in women?

What research and autistic adults describe most consistently in this presentation is a social performance that works. Eye contact is managed rather than natural. Conversation is scripted from observation. Difficulty appears as exhaustion after social contact rather than as visible struggle during it, and the sensory load, the routines and the intense interests are present but kept private because they have been read as odd before. The signs are real. What makes them hard to see is that they are being actively hidden, and the hiding is the skill.

Autism Speaks, UCLA Health and the NHS all publish versions of this list, and a list is where most people start. The thing a list cannot tell you is the timing, which is the part I want to spend this essay on.

Each of those behaviors is observed by someone else, during the performance. The cost is paid afterward, alone. That split is the whole reason forty years can go by. A reference set built on behavior that is visible in the room will not record a cost that lands in the car on the way home.

The second thing a list cannot tell you is that the better the performance, the weaker the signal. A person who has been reading rooms since childhood is not producing the behaviors the criteria describe. They are producing the opposite of them, deliberately, at a price no review cycle has an instrument for. I wrote about what the price consists of in masking autism and about what the label grades in high-masking autism.

Why competence reads as the absence of difficulty

Competence stops the question. If the work lands, the meeting goes well and the client is happy, there is no observable problem for anyone to explain, and nothing in a review cycle is pointed at a cost paid outside working hours.

The research dismantles the assumption underneath this more directly than most people expect. Alvares and colleagues found that intelligence is an imprecise predictor of functional abilities at diagnosis (Autism, The misnomer of “high functioning autism”). Being able to do the hard thing tells you very little about what doing it costs. The whole functioning vocabulary rests on the opposite assumption, which is why I have written about that label separately.

From the other side of the table, where I sat for most of my career, the loop is plain and no one in it is being careless. A person who delivers gets the next piece of work. The loop is reading the only signal it collects.

Now put a forty-year career through that loop. Every successful performance raises the expectation for the next one. The support stays at zero, because no need was ever visible. The recovery window has to grow to cover the gap, and it grows in private. By the time anything forces the question, there are four decades of evidence on file that this person is fine.

What are the signs of level 1 autism in women?

Level 1 is the support-needs label the current diagnostic manual assigns when support requirements are described as the lowest of three tiers. In women it tends to describe someone who is employed, often successfully, managing social contact through learned strategy rather than ease, and depleted by it in ways that are not visible at work. The level describes how much support a clinician recorded at one assessment. It does not describe how much the person is doing to hold the presentation together.

That last distinction is the one worth carrying. A support level is a snapshot of an observation, taken during the performance, by the same instrument that missed the person for decades. It is useful for allocating services. It is a poor description of a life.

What the research can measure, and the one thing it cannot

The mechanism now has a measurement behind it. Hull and colleagues built and validated a camouflaging questionnaire in 2019 that separates the work into three jobs: compensating for difficulty, masking the signs of it, and assimilating into the group well enough to pass (Journal of Autism and Developmental Disorders, Development and Validation of the CAT-Q). Three tasks, run simultaneously, on top of whatever work the room was actually for.

This month a systematic review aggregated what nine studies have found about where that lands. Cuni and colleagues reviewed work published between 2015 and 2025 on masking and mental health in autistic women, a combined sample of more than fifteen hundred women across seven countries (Journal of Autism and Developmental Disorders, Autistic Masking and Mental Health Outcomes in Autistic Women). Masking was consistently associated with anxiety and with depression, and was an independent predictor of suicidality, with the assimilation component specifically predicting suicidal ideation. Two things about that review are not optional to repeat. Its figures come from different studies with different designs, so there is no single effect size to quote. And it required a formal diagnosis to be included, which means it screened out the undiagnosed and late-recognized population. The people this essay is about are not in it.

The review also says, in one sentence, the most useful thing in it: no study assessed autistic burnout with validated instruments. Masking to anxiety, depression and suicidality is now quantified. Masking to burnout is the step everyone assumes and nobody has instrumented. I believe the link, and I will not describe it as measured, because it isn’t.

I am also reading this as an operator rather than as a researcher. The authors’ recommendation is clinical screening. Taking their finding and turning it toward hiring loops and review cycles is my argument, not theirs.

What are the 12 signs of autism in adults?

There is no canonical list of twelve. The number circulates because listicles need a count, and different pages assign it to different behaviors. No diagnostic manual publishes a twelve-item set for adults, and anyone presenting one as official has made it up.

What is real is the question underneath the search. Someone is looking for a threshold, a point at which the pattern stops being a personality and becomes a thing with a name. The honest answer is that no list produces that threshold, because the criteria were written to be applied by an assessor across a developmental history rather than scored at a kitchen table. A list can tell you whether to take the question seriously. It cannot answer it.

Why this profile is identified late, or never

Three mechanisms stack, and each one is individually reasonable.

The reference set is the first. A presentation outside it does not fail the criteria so much as never reach them, because the referral is the step that never happens. A child who is quiet, compliant and academically fine does not generate a referral, and the adult version of that child does not either.

The second is the labels that get applied instead. Something is observed, because something is always observed, and it gets recorded under whatever vocabulary the setting has. Anxious. Shy. Sensitive. Difficult. Too much. Not a team player. Each of those is a real observation attached to the wrong mechanism, and each one closes the file. I have written for executives about what those labels cost the companies that record them.

The third is the performance itself, which strengthens over time rather than lapsing. Suppressing the visible regulating behavior is part of that work, and I have written about what suppression actually costs in stimming. A ten-year-old gets praised for being easy. A twenty-five-year-old gets promoted for reading the room. A forty-five-year-old has forty-five years of positive feedback on the performance and no vocabulary for the price. The system is not hiding anything. It is rewarding the thing that makes the question invisible, which is a more durable arrangement than concealment.

What recognition looks like from the inside

The thing I did not expect is how little changed. My capacity stayed the same. The work stayed the same. Nothing about the previous four decades was retroactively fixed, and I have no interest in writing a version where it was.

What changed is which question I ask. Before, a lost day was evidence about me, and the evidence was always the same and always unflattering. After, a lost day is evidence about the day. That is a reattribution rather than a recovery, and it is the whole of what recognition did. I wrote about the arrival itself in late autism diagnosis.

The second change is that the environment becomes a variable you can actually move. For forty years the only adjustable thing in the system was me, because I was the only part of it anyone had identified. Recognition adds the room, the schedule, the meeting format and the amount of social contact in a week to the list of things that can be changed. Some of those are changeable at work and some are not, and being honest about which is which matters more than any of the rest of it.

What this does not tell you

I lack a general solution and I am suspicious of anyone selling one. I cannot tell you whether you are autistic. I cannot tell you whether to pursue an assessment, which is a decision with real costs attached that vary enormously by where you live and what you do for a living. I cannot tell you whether to disclose at work, because the answer depends on the environment, and some environments punish it accurately.

What I can tell you is what the instrument was built from, why a working performance reads as no finding, and where the cost goes when nothing collects it.

The understanding of this presentation did not come from me and it did not come from the clinical literature first. Autistic women described it, in their own writing and in each other’s company, years before there was a questionnaire with a validated score attached. The research caught up. The vantage point here is mine. The insight is theirs.

Forty years is not how long it took the person to notice. It is how long the instrument looked at her and recorded nothing.

Sources

Peter Allen Mann is the author of The Perception Revolution (Fast Company Press, 2026). He spent thirty-seven years in leadership, including four years as a Navy Surface Warfare Officer and two companies founded and run, and was diagnosed autistic at fifty-five. He writes for executives and system designers at SHIELD Perception.