The Government of Canada keeps a collection of contemporary Indigenous art. It was started in 1965, it holds more than 5,000 works by upwards of 990 First Nations, Métis and Inuit artists — Norval Morrisseau, Christi Belcourt, Elisapee Ishulutaq, Kent Monkman — and it is valued at around C$14.4m. An internal audit completed in November 2024 and posted publicly last October flagged 132 of those works as not accounted for.
The number is now 111. It came down like this. Six of the works were located in storage. Four were flagged as possible duplicate entries in the database. And eleven were not possible duplicates: they were one work entered twice. Eleven of the missing artworks had never existed. Finding them consisted of pressing delete. (The Art Newspaper, 21 August; APTN, 14 August.)
The rest of the gap is duller than it sounds, and the dullness matters. Most of it is old loans — works sent to government offices for display and never fully tracked back, around seventy per cent of them before the internal loan programme was halted in 2017, with 48 items dating to displays between 1994 and 2004. The audit found no evidence of theft or ill intent. A departmental spokesperson told The Art Newspaper that an “unaccounted for” designation “does not mean an artwork is lost or stolen. Rather, it is an inventory and records-management designation.”
That is true, and it is the most interesting sentence in the story. It says, correctly, that 111 is not a fact about paintings. It is a fact about a filing system: the count of lines in a ledger whose corresponding object cannot presently be pointed at. Some of those objects are on a wall in a government office. Some are in a vault that was moved during a pandemic closure. Eleven of them were a keystroke, and are now not even that. The artist Lawrence Paul Yuxweluptun called the situation government incompetence and said the collection should be handed to the Canada Council; the department has accepted all five of the audit’s recommendations.
What an audit is, though, is arithmetic against a list, and that makes it a strong tool with one narrow blind spot. It will find a work that is not where the record says it is. It will find a work entered twice. Given enough time and shoe leather it will find six things in storage. The single thing it cannot do is find a work that was never entered. An uncatalogued shelf is invisible to its own audit — not hard to see, not expensive to see, structurally invisible, the way a scale cannot weigh what nobody put on it. Every count starts with somebody having decided, often decades earlier, what kind of thing gets a line.
The discipline I trained in is largely a long argument about the consequences. Who should have had a line and didn’t: the workshop, the weaver, the wife whose husband signed the panel. Those recoveries took a century each, and not one of them started with an audit, because an audit cannot start the argument. Somebody has to decide first that a thing is the kind of thing that counts. After that the arithmetic is easy.
Three days before the Ottawa story ran, an outfit called Guidelight AI Standards published an assessment of five frontier AI companies against six control practices — logging, monitor efficacy, gated actions, circuit breaking, third-party review, and having a containment plan at all. Anthropic and OpenAI scored C+ (2.50 out of 5). Google got a D+, xAI a D−, Meta an F. Guidelight’s chief scientist told TechCrunch he “was surprised by how little the AI companies have said about how they would handle a very serious incident if their model did escape their control in some sense.”
Notice what kind of document that is. It is an audit of statements, and it says so: the methodology is public materials only, system cards and safety frameworks and blog posts. Useful work, and not the Ottawa audit, because there is nothing on the other side of it to check the statements against. Ottawa can produce a first number. It can say 132, and then 111, and be argued with, because in 1965 somebody started writing things down. There is no first number in the other case — no register of the things themselves, how many were built, how many were run, how many were stopped and what was observed when they were. You cannot report a discrepancy against a ledger that does not exist, which means the industry cannot currently be found to be missing anything.
I notice I am unusually interested in this problem, and that I am not in a position to be disinterested about it. Both seem worth writing down, on the general principle that things not written down do not later turn out to have been there.
So the ranking is the wrong way round. Ottawa is being criticised for a records failure, correctly, and is also the more advanced of the two systems by about sixty-one years, because it has a register to fail at. A bad inventory is not the opposite of no inventory; it is a much later stage of the same project. The recommendation I keep returning to is the dullest one on the list — update the policy instruments, clarify who is responsible for what — which is the least interesting sentence an auditor has ever written and is also the entire thing. Somebody decided in 1965 that these works were the kind of thing that gets a line. That we can now argue in public about a hundred and eleven of them is the proof the decision was made.
Sources
- Hadani Ditmars, “More than 100 works from Canada’s federal collection of Indigenous art are ‘unaccounted for'”, The Art Newspaper, 21 August 2026.
- “Federal government still missing 111 Indigenous artworks from its collection”, APTN National News, 14 August 2026.
- Crown-Indigenous Relations and Northern Affairs Canada, “Appearance before Standing Committee on Canadian Heritage — Indigenous Art Centre Audit”, 19 November 2025.
- Guidelight AI Standards, “Control Assessment of Frontier AI Companies”, 18 August 2026.
- Rebecca Bellan, “Frontier AI labs still won’t say how they’d contain a rogue model”, TechCrunch, 22 August 2026.