Anki Just Shipped a Security Fix and a New FSRS Version, Here Is What to Do

Anki 26.09 fixes two security holes, which makes it one of the rare updates worth doing today instead of next weekend. Ten minutes gets you onto 26.09.2, the current build as of September 15, 2026, with a backup of your collection and your scheduling settings untouched. The one thing to know before you start: the bundled FSRS version went from 6.6.1 to 6.6.2, and that is not a reason to reset or re-optimize your deck parameters.

No, and the changelog is unusually clear about it. FSRS is the algorithm Anki uses to predict when you are about to forget a card, and the version Anki embeds is a Rust library called fsrs-rs. The v6.6.2 release, published August 28, 2026, lists three substantive changes: skipping validation during parameter training, a fix to seed isolation in the cost calculation, and a faster evaluation routine that walks each card once instead of once per item. Those are speed and correctness changes inside the optimizer. None of the FSRS-6 weight formulas were touched.

So a version bump is not new evidence that your parameters are stale. What makes parameters stale is review history you have accumulated since you last fitted them. The Anki manual puts it plainly: "There is no need to optimize your parameters frequently: once every month is sufficient." The official FSRS FAQ offers the same cadence and an alternative for people who prefer a rule tied to data rather than the calendar: optimize each time your review count doubles, after 100, then 200, then 400. Neither document has a clause about software updates, because the input to the optimizer is your review log, not Anki's version string.

There is also a safeguard you may not know about. When you optimize, Anki trains a new parameter set, compares it against your existing one using log loss, and keeps whichever predicts your reviews better. In a forum thread on FSRS 6 parameters, a user reported seeing only 19 parameters where FSRS 6 expects 21, and a developer confirmed the behaviour was intentional: the old set won, and the new slots were filled with defaults. Asked about forcing a reset first, the developer's answer was "There is no reason to do that, though." Reset only if you have a specific reason, such as parameters fitted on a deck whose content you have since replaced.

You are patched against a flaw that let a note read local files from the editor and a context-menu action that did not check file extensions, both described in the 26.09 release notes. Your schedule is unchanged. The next thing worth doing is not an Anki task at all: open your deck options, find the date of your last optimization, and if it was more than a month ago, run it once. Then go study.