A critical vulnerability in Apple's Screen Sharing service let unauthenticated attackers take over a targeted Mac over the network — reading and writing files as root with no password, no username, and no user interaction. In the researchers' tests, a single pipelined connection could drop a root shell in under 60 seconds.
Apple quietly fixed the flaw in macOS 26.6, released July 27, but the story of how it surfaced has become one of the year's more contentious disclosures.
The bug sits in screensharingd, the background daemon behind macOS Screen Sharing. Its authentication routine mishandled oversized data frames: when the daemon received a frame flagged as "too large," it returned the leftover success code from a previous read instead of an error.
That single stale value convinced the daemon the login had succeeded — so it skipped the entire password check, key exchange, and encryption setup, and dropped the attacker into a fully privileged session running in plain text. Screen Sharing normally wraps traffic in ChaCha20-Poly1305 encryption; here, none of it was ever switched on.
From inside that unauthenticated session, Apple's built-in file-copy protocol handed over arbitrary file read and write as root. The read primitive alone exposes the whole disk — home folders, logs, app data, and configuration. The write primitive lets an attacker drop files wherever the system allows.
On a default Mac, System Integrity Protection (SIP, the kernel-level guard on system files) blocks the most direct routes to code execution, but researchers still demonstrated a SIP-compatible path by planting a passwordless sudo rule and a shell-startup hook that fires the next time anyone opens a terminal.
The advisory flags a second, independent weakness too: the daemon's SRP (Secure Remote Password) login check fails to reject certain malformed values that can force the session's shared secret to a predictable result — a separate password-bypass that wasn't even needed for the main attack.
Veteran macOS reverse engineer fG! (of reverse.put.as) says he independently found the same pre-auth bug long ago and argues that the team that reported it to Apple — reportedly leaning on an AI-assisted, GPT-driven discovery workflow — logged it as a lesser denial-of-service crash and missed its true severity. "It's a pre-auth, stupid!" he wrote, taking aim at both Apple's patching pace and the industry's reliance on LLMs to hunt bugs.
A second researcher then reverse-engineered his deliberately limited proof-of-concept into a full exploit chain — one that even self-destructs, wiping its own planted files once the connection lands. That chain was corrected within a day: its authors first claimed reliable code execution on a stock Mac, then walked it back after realising their test machine had SIP switched off.
The bottom line for users is blunt. SIP is not a fix — on a default, fully-updated-but-pre-26.6 Mac, an attacker could still read the entire disk and write to sensitive locations without ever logging in. Any Mac on macOS 26.5 or earlier with Screen Sharing enabled is exposed, and unpatched older systems on home and corporate networks are the softest targets.
How to protect yourself:
- Update to macOS 26.6 right away.
- If you can't patch, turn off Screen Sharing under System Settings → General → Sharing.
- Keep SIP enabled — it won't stop the break-in, but it sharply limits the damage.