SlowMist puts 14 iOS versions inside one Safari key-theft chain

A full chain is several separate bugs stacked in order, and this one makes four moves after the browser foothold: past Pointer Authentication Codes, out of the sandbox, up to kernel level, into the Keychain ^[2][13]. Each move needs its own fix. Apple has patched several of the vulnerabilities in the chain through iOS 26.3, according to the report, while the affected range SlowMist published ends at 26.5, two point releases higher ^[4][1][12]. The instruction to update iOS, which Ledger's Charles Guillemet gave on September 21, is the best available step, and whether it closes the chain is undemonstrated ^[3].

Fourteen major versions sit inside the stated range, iOS 13 through 26 ^[11][1]. A range that wide leaves the number of handsets actually reached open, and SlowMist published no CVE number, victim count or loss figure ^[14].

Two people carry this on the record: SlowMist's CISO, who publishes as 23pds and issued the alert on September 19, and the chief technology officer of Ledger, who amplified it two days later ^[3][1]. The remedy the article settles on is architectural: keys held on a dedicated physical device that never touches a network, out of reach of a Safari bug ^[5].

For a holder who suspects exposure, the recommended fix is not a password reset. Researchers advise regenerating keys entirely on a clean device and moving funds to a new wallet address ^[6]. A stolen private key has no reset at all; whoever holds it holds the funds ^[7].

This is the second iOS episode aimed at crypto users in 2026, by cryptobriefing's count ^[9]. In March, researchers flagged an exploit kit called DarkSword that chained six separate vulnerabilities against iPhones on iOS 18.x variants and went after Coinbase and MetaMask ^[8]. The March kit used six bugs; the September chain uses four escalation steps ^[8][13].

I would treat the published version range as the operative fact here and a single firm's confirmation as thin evidence. The asymmetry still favours rotating keys: being wrong about waiting costs the balance ^[7], while being wrong about rotating costs transaction fees and the work of standing up a new address ^[6]. The more interesting decision sits behind it: whether a general-purpose phone is a custody venue at all, given that a software wallet shares its attack surface with every app, webpage and piece of code on the device ^[10]. Two things would change my mind. An Apple security note naming the CVEs fixed in 26.4 or 26.5 would collapse the gap between the patch line and the top of the affected range ^[12], or a second firm publishing victims and on-chain losses would move this from a version range to a measured event.