Australia warns of 'unprecedented' AI threat

Australia's Signals Directorate has surveyed the past twelve months and reached for the word "unprecedented," a word now worked so hard it ought to be paid overtime and given a pension. Foreign governments, the agency says, are "turning AI loose" on computer networks and critical infrastructure — a phrase that summons the image of a farmer opening a gate, except the gate is a water treatment plant and nobody has read the terms. France 24 carries the warning without naming a single one of those foreign governments, which is either operational discretion or the diplomatic equivalent of "we'd rather not, and you can't make us."

The figures: more than 4,400 warnings about potentially malicious activity in the past year, up 80 percent on the prior twelve months. That is roughly a dozen a day, every day, Christmas included. "Potentially malicious" is the phrase doing the real lifting — the agency is flagging things that might be an attack, which is a fire alarm that also goes off when someone makes toast. And the 80 percent arrives with no absolute baseline, so we are invited to admire the slope of a graph whose earlier data point has been left in the drawer, presumably for brevity.

None of this is wrong, exactly. The direction of travel is legible and it terminates in a substation. But note the remedy on offer: more warnings, at a rate 80 percent higher than last year, dispatched into an inbox where a sysadmin on a Tuesday decides whether this is the real one. "Unprecedented" is, at bottom, the word you use when the honest sentence is "we would like somebody to fund this properly, and it isn't going to be us." The adversaries remain unnamed, the AI remains unspecified, and the number went up by a percentage whose other half is missing. Cheers to being alerted.