This question usually arrives attached to a hope, and the honest answer does not entirely satisfy it. Attachment is not a personality tattoo and it is not a setting you can flip. It sits in the awkward middle that most real psychological findings sit in: stable enough that it predicts things about you a decade out, unstable enough that a good share of people are somewhere else when you measure them again.
What follows is what the evidence actually supports, including the parts that are inconvenient for the story.
What the research actually shows
Stability is moderate, not high. Across long-term studies the correlations between attachment security measured early and measured much later are real but far from unity — the kind of number that tells you the trait is doing something and also that a substantial minority of people have moved. That is the single most useful sentence about this whole topic, and it cuts both ways: nobody should be told they are stuck, and nobody should be told it is a matter of will. The same split runs through personality generally — genes, environment and the story you tell about yourself each get a share, and none of the three gets all of it.
Security is partly relationship-specific. You do not carry one number around. You have a general tendency plus a reading for each important relationship, and the relationship-level one moves faster. This is why people report feeling like a different person with a steady partner — that is real, it is measurable, and it is the most reliable route by which anything changes at all. The size of it is worth stating: across a year, attachment measured toward a parent barely moves, while attachment measured toward a romantic partner moves a great deal, which is the subject of why security arrives with one person and not another.
Change is usually slow and usually relational. The two things that show up in the literature as actually associated with movement toward security are a long-term relationship with a secure partner and a long-term therapeutic relationship. Both are years, not weeks. Neither is a technique. What they have in common is repeated, unremarkable evidence that the expectation was wrong — which is the only thing that updates an expectation formed before you had language.
Life goes the other way too. Attachment moves toward insecurity as well. Loss, betrayal, a long period of instability — these shift scores, and that is part of what "moderate stability" means. A model that only allows improvement is not describing the same thing the studies are.
Earned secure, and the asterisk on it
The term comes from a specific observation: some adults describe difficult childhoods and nevertheless talk about them coherently, without idealising or still being inside them, and by the standards of the Adult Attachment Interview that coherence is what security is. Those people were labelled earned secure, and the label travelled fast because it is a good story.
The asterisk is that the label was originally applied retrospectively — from the adult's own account of their childhood — and when researchers finally had a cohort with prospectively collected childhood data to check it against, the picture got complicated. The cohort in question is the Minnesota Longitudinal Study of Risk and Adaptation: children of low-income first-time mothers in Minneapolis, recruited in the 1970s and followed from before birth. When Roisman and colleagues compared retrospectively defined earned-secure adults with the observational records of their actual early care, the harsh childhoods those adults described were not clearly there in the prospective data — and those adults reported more depressive symptoms than continuously secure ones. The prospectively defined group, the people whose early care really had been poor and who were secure as adults, turned out to be a different and smaller set of people, and they did not show the same elevation.
Two things follow, and they are both worth holding. The first is that earned security is a real phenomenon and not marketing: there are people whose measured early care was poor and who are secure adults. The second is that the term is used loosely, that the retrospective and prospective versions pick out different people, and that the strongest evidence comes from one American cohort recruited from low-income families fifty years ago. How far that generalises is genuinely open. The entry on earned secure attachment has the studies laid out properly; attachment theory has the frame.
What actually moves it
- A steady relationship, given years. Not a rescuer and not a project — someone whose behaviour is boringly consistent, long enough for the boring consistency to stop being surprising. This is the one with the most support behind it and the least control attached to it.
- Therapy, for the same reason. A relationship with reliable conditions, in which the expectation gets disconfirmed often enough to update. The modality matters less than the duration.
- Disconfirming evidence you cannot argue with. Small, specific, repeated. Saying the thing you would normally manage alone, and noticing that nothing happened. A hundred of those do more than any insight.
- What does not move it: deciding to be secure, reading about attachment, or identifying strongly with your style. The last one is worth flagging — treating "I'm an anxious attacher" as an identity tends to make it more permanent, because it converts a behaviour you could change into a category you belong to.
What it depends on
How much room there is to move depends on where you are starting, and specifically on which dial is high. Anxiety tends to be the more responsive of the two, because it is loud enough to be noticed and worked with directly. Avoidance is quieter and slower, and the first difficulty with it is that the person rarely experiences a problem to solve.
The practical version of all this is unglamorous: get a reading now, do the unremarkable things, and take another reading in a year. Change at this scale is not something you can feel happening — it is only visible as a difference between two measurements. Without a baseline, a year of work leaves you with the same question you started with.