When data must be deleted, a copy often survives, or nobody can prove it is gone. This map shows what followed, country by country.
USD 8,436,379,254
in fines, settlements and damages across 429 cases on this map.
Failed to delete USD 2,881,236,553 · Had to delete USD 743,672,484 · Settled, not proven USD 4,806,848,023
A fine is only one cost. Organisations that kept data too long also paid to respond to the breach, to protect customers and to repair systems. This view lists the amounts they reported. None of these amounts is in the total at the top of the page.
Each row is an event that involved old data. The fourth column says how we know. Regulator finding: a regulator found that data was kept too long. Old data involved: the organisation or a regulator said that former customers or old records were in the event; it does not mean the data was kept unlawfully. One row rests on a regulator's allegation that was settled without an admission, and says so. No row says that old data caused the loss.
Each amount is the whole loss the company or source reported for the event, not only the part linked to old data. Read on 9 November 2026. Australian dollar and Korean won amounts are converted at about the rates this map uses for its counted cases (AUD 1 = USD 0.70; CAD 1 = USD 0.71; KRW 1,000 = USD 0.76).
Where a breach published the split, this share of the exposed records should already have been deleted: AT&T about 90% (former customers), T-Mobile 84%, Latitude 77%, Fairview Park 64%, Cyberport 40%, Golfzon 17%, McDonald's Korea 16%, Coupang 7% or less. All are on this map. A breach cannot expose a record that is gone.
A fall in the share price. Customers who leave. Higher insurance premiums. Management time. The cost to review old data in a lawsuit. In defence: the harm when classified or export-controlled data is kept and leaks, and the cost when allies share less because they cannot trust that data is destroyed. No source on this map gives a figure for these that is tied to data kept too long. So the amounts in this view are a floor.
In part. The projection has a line for breach cost from data that should already be gone: USD 12.5bn for 2026. The amounts in this view are costs of that kind. The model estimates the category as a whole; it does not add these rows. It has no line for share price falls, lost customers, insurance or management time.
This view is a model, not decided cases. It estimates two figures for each year, in US dollars.
What data that should have been deleted costs organisations each year.
USD bn a year, log scale. Circles mark dated steps.
What organisations pay each year to delete data and to show that it is gone.
Defence, not in this line: the US Department of Defense puts the yearly cost of its CMMC cyber rules for contractors at USD 4.2bn. Destroying data is one of those rules.
USD bn a year, log scale. Circles mark dated steps.
Exposure is the problem. Spend is the money put against it. The gap is cost that organisations carry and have not paid to remove.
Do not add the two figures. Spend partly offsets exposure.
So exposure is a floor. The rows in Losses beyond fines are examples of breach cost; the model does not add them one by one.
| Compound annual growth rate | 2026 to 2030 | 2030 to 2035 | 2026 to 2035 |
|---|---|---|---|
| Exposure | 54% | 39% | 45%* |
| Spend | 34% | 49% | 42% |
What the rate means. A compound annual growth rate is the steady yearly rise that would take the 2026 figure to the later one. At 45% a year, exposure is 29 times larger in 2035 than in 2026. Exposure grows fastest before 2030, because three dated steps land by then. Spend grows fastest after 2030, because buying lags enforcement by about two years. For scale, measured series today: breach cost per breach rises 12% a year, privacy software 23%, data created 24%, deletion requests about 45% to 60%, frontier AI training compute 400%.
The shape. A staircase on a curve that bends upward. The steps are dated rules that create a new kind of invoice. The curve underneath is AI: each new agent is another place data is copied to, and agents double what they can do about every seven months. Copies grow faster each year, so the underlying rate rises each year, from 12% in 2026 to 40% by 2032, where the model caps it. Spend follows exposure two years later, because buying lags enforcement. If a rule is delayed or rolled back, that year flattens and the curve resumes.
* Three questions readers ask. Is 45% a year believable? Without the acceleration the model gives USD 34bn in 2035; without the steps, USD 75bn. The half-trillion needs both, and both have dates. For a year you can check, 2028 gives USD 39bn exposure and USD 16bn spend on rules already passed. Does cost follow copies? Where breaches published a split, most of the exposed data should already have been gone: AT&T 90%, T-Mobile 84%. Every agent, cache and backup is one more copy. How big is each step? The dates are law; the size is judgement. A 1.3 step is smaller than 2026, when decided penalties on this map went from USD 70m to USD 3bn. The page sizes the category, not any company's share. The gap between the two lines is proof nobody has priced.
Modelled 31 October 2026. Intelligence, not legal or investment advice.
Click a bar to see one year.
Click a row to filter the map.
Click a case to read it.
1,730 cases and laws from 239 countries and territories. Every entry links to its source.
Some entries are not tied to one country: EU-wide rules and guidance, and studies that cover many countries.
This page has three views. Fines and settlements: cases that a regulator or court has decided, or that were settled; these make the total at the top. Losses beyond fines: other costs that organisations reported; these are not in the total. Projection 2026 to 2035: a model of the yearly cost; it is an estimate, not decided cases.
Each case shows a duty to delete that was missed, or that could not be proven. It does not necessarily show that any product would have prevented the breach or the fine. Fines are the total penalty; in most breach cases the deletion failure was one finding among several, and the entry says so where we can see it. "Why it matters" is our reading, not the regulator's. Where an order is cited, it shows a duty; it does not necessarily show the deletion was carried out, unless the entry says it was checked. Entries marked Secondary rest on news, law firm or commentary sources; check the original before quoting.
"Law, no case yet" means we found no published case, so the entry cites the law that creates the duty. "No general data protection law" is taken from the source cited for that place; sector, constitutional or contract duties may still apply.
Dots mark small states and territories that have a case. Small places with a law and no case appear when you zoom in; all are in the list below the map. Overseas departments of France sit inside France. Places follow the Natural Earth 1:50m country set, so Gibraltar, Tokelau and the Sovereign Base Areas are not separate shapes.
Send us the decision and a link to the source. We check every one against its source before it is added to the map and the counter.
Send us a fineThis map is our best research from public sources, mainly regulator and court decisions, laws, official guidance and published research. We check each entry against its source before it goes live. The map is not complete, sources can change, and errors are possible. Some decisions are under appeal or not yet final; the entry says so where we know it. Amounts are rounded and converted to US dollars as explained above. Where an entry cites a lawsuit, an investigation or a reported incident, it is not a finding that the law was broken. This map is for information only and is not legal advice. To report an error or a missing case, email info@together-alone.com.
This total is a floor. It counts only cases a regulator or court has already decided and published; breaches avoided because data was never kept cannot appear in any dataset. Within each case the amount is the whole penalty unless the decision splits it by breach, so not every dollar is for deletion.
Together Alone Ventures OÜ trading as Zombie Delete · Company number 17288376 · Tartu mnt 67/1-13b, Harju maakond · 10115, Tallinn, Estonia · © 2026 Together Alone Ventures OÜ