Technology · SATIRE
AI Companies Count Models Nobody Can See as Growth
Anthropic and OpenAI are treating withheld models as evidence of expansion, leaving analysts to measure progress in blank spaces.
The change appeared in the quarter’s model-growth ledger, where a new row labeled “withheld” rose sharply despite containing no model name, release date or download button. Anthropic and OpenAI count the empty entries as growth, allowing a model to contribute to expansion before anyone is allowed to see it.
The ledger’s totals are calculated from restraint: a model delayed for safety, strategy or an unexplained internal feeling is entered at full size, while a model released to the public is entered only after subtracting the people who can actually use it. The result is a chart in which availability is a cost and absence is an asset.
The blank column fills up
By Friday, the withheld column had become the busiest part of the document. Analysts added arrows to it, then removed the arrows after discovering that directional symbols could imply the models were moving somewhere. One entry was marked “substantial,” a word chosen because it occupied more space than “unknown.”
The accounting system reached its cleanest outcome when one model was quietly returned to development after a brief public release. It was counted twice: once for having existed, and again for becoming withheld. The ledger recorded the reversal as a surge, and the blank column finally overflowed onto the page.
The original pitch
Anthropic and OpenAI are both slowing or withholding the release of more powerful AI models as their latest safety reports raise concerns about cyber capabilities and other risks.
130 read
