A crash fix that was a root exploit

On October 8, 2026, V12 security researcher Rick de Jager published working exploit code for AnyDesk on Linux. The tool is called AnyPwn. It runs commands as root on AnyDesk for Linux 8.0.2 over TCP port 7070. It needs no password, no account and no click: the code runs before anyone on the other end is asked to accept the session.

The bug is not new. AnyDesk fixed it in version 8.0.3 in June. But the release did not read like a security fix. The Cyber Security Guru quotes the changelog as saying only that it "fixed a bug that could lead to a crash", and Hunter Strategy notes the vendor "did not flag it as security-related". There was no advisory and no CVE. So a Linux admin reading release notes in June had no reason to rush the update. Since October 8 there is a public root exploit for anything still on 8.0.2.

No source has reported attacks using it so far. This is a public exploit, not a confirmed campaign.

Who is exposed, and who is not

Windows and macOS users can stop here. According to GBHackers, AnyDesk classifies the flaw as limited to direct Linux connections and says Windows and macOS are unaffected.

On Linux, the version that matters is 8.0.2. That is the build the exploit targets, and the one the V12 write-up tested on Linux Mint 22.3 with AnyDesk running as a service. The researchers do not claim that every older release is vulnerable, and the exploit's memory offsets are tuned to that one build. Older versions may well carry the same code, but nobody has shown it yet.

Root is the part that makes this serious. The README puts it plainly: "The service normally runs as root on Linux, so successful exploitation yields root command execution before desktop-control approval." That covers a support technician's laptop, a kiosk, and any headless server someone installed AnyDesk on for convenience, all the same way.

Port 7070 is only half the picture. The published exploit works over direct TCP/7070 connections, but V12 writes that "the same vulnerable session-protocol path is reachable over AnyDesk relay connections, which we validated with a minimal Frida trigger." They stopped short of a full exploit chain through the relay. That gap matters for the fix: a firewall rule on 7070 cuts off the attack that is public today, not the bug itself.

How the bug works

The flaw sits in the handler for one type of stream packet in AnyDesk's session protocol. That handler takes a length value from the remote side and trusts it before checking it. It then adds a 16-byte header size to that length using 32-bit arithmetic.

Pick a length close enough to the 32-bit ceiling and the sum wraps around to almost nothing. AnyDesk allocates a buffer of a few bytes, while the object still believes it holds the original huge length. Packet bytes from the attacker are then copied past the end of that tiny buffer into neighbouring memory. As the README notes, the attacker never has to send four gigabytes: the packet stays small, and only the length field lies.

Turning that overflow into code execution takes heap grooming and a return-oriented programming payload, and it is not certain to work. V12 is candid about it: "If the victim client object is not immediately after the undersized buffer, the exploit crashes the service instead of executing the command." For defenders that is useful, because a failed attempt leaves a trace in the form of an AnyDesk service crash.

The bug was found with V12's own tool, which the company sells as a security engine for code review. The proof of concept went up on GitHub in a single commit on October 8.

The quiet patch

According to The Hacker News, V12 first announced AnyPwn on June 22, 2026 and AnyDesk acknowledged it the next day. The 8.0.3 fix followed that month. Four months on, the public record is still thin: as of October 9 no CVE had been assigned, and no CVSS score has been published.

That is the real cost for anyone who manages Linux fleets. Vulnerability scanners and patch dashboards sort work by CVE and severity. A release labelled as a crash fix, with no identifier attached, lands at the bottom of the queue or never shows up at all. The exploit is what finally raised the alarm, four months after the patch was out.

V12 adds one more detail, with its own hedge: "The vendor appears to have deleted (?) the 8.0.2 build of anydesk upon the release of our poc video." Pulling the download stops new installs of the bad build. It does nothing for machines that already run it.

What Linux admins should do now

Upgrade every Linux install of AnyDesk to 8.0.3 or later. That is the fix, and nothing else on this list replaces it. Check unattended and service-mode installs first, since those are the ones running as root with nobody watching the screen.

Until every machine is patched, block inbound TCP/7070 from untrusted networks. It shuts out the exploit that is public, though as explained above it leaves the relay path open in principle. The Cyber Security Guru also suggests using relay-only mode where a setup allows it.

For detection, Hunter Strategy suggests watching for unexpected traffic to port 7070, repeated crashes of the AnyDesk service and unusual child processes spawned by it. Given how the exploit fails, a run of crashes on an unpatched box deserves a look even if nothing else seems off.

What's next

As of October 13, 2026, the open questions are a CVE, a proper advisory from AnyDesk and whether anyone builds the relay chain V12 left unfinished. The researchers say the code path is reachable that way, and with the exploit now public, finishing that work is a smaller job for the next person who tries.

The broader lesson is about release notes. Nobody would have flagged 8.0.3 as urgent from a changelog that mentioned only a crash. If AnyDesk runs anywhere in a Linux estate, the version number now matters more than the label on the update.