Key takeaway: CVE-2026-45185 — a use-after-free vulnerability in Exim's BDAT (binary data transmission) message body parsing — allows unauthenticated remote code execution on one of the world's most widely deployed mail transfer agents. Affects Exim versions 4.97 through 4.99.2 when built with GnuTLS. Patch to 4.99.3 immediately.
What Happened
On May 12, 2026, the Exim development team released version 4.99.3 to address CVE-2026-45185, a use-after-free vulnerability in the BDAT message body parsing path. The flaw, nicknamed "Dead.Letter," affects all Exim builds using the GnuTLS library for TLS connections — which is the default on most Linux distributions. The vulnerability can be triggered by a remote, unauthenticated attacker sending a specially crafted email.
Exim is the default MTA on Debian, Ubuntu, and numerous other Linux distributions. It's estimated to handle a significant portion of the world's email traffic. The vulnerability was discovered by the XBOW research team, who notably used AI assistance to help develop the working exploit — making this the second confirmed case of AI-assisted zero-day development in the same week.
Technical Analysis
The vulnerability is a use-after-free in Exim's BDAT handling code when TLS is managed by GnuTLS. The attack works when a client sends a TLS close_notify alert before the BDAT body transfer completes, then follows up with a final byte in cleartext on the same TCP connection. This sequence causes Exim to write into a memory buffer that has already been freed, resulting in heap corruption.
The affected versions are Exim 4.97 through 4.99.2 when compiled with GnuTLS support. Builds using OpenSSL are not affected. The vulnerability is remotely reachable — meaning any internet-facing Exim instance accepting SMTP connections is potentially exploitable without authentication.
The XBOW team demonstrated that the heap corruption is exploitable for arbitrary code execution. Combined with the unauthenticated nature of the attack, this makes CVE-2026-45185 a critical-severity vulnerability. While CVSS scoring was not yet published at the time of writing, the combination of remote, unauthenticated access and code execution potential places this firmly in the 9.0+ range.
Who's Affected
Exim is one of the most widely deployed mail transfer agents in the world. It ships as the default MTA on Debian, Ubuntu, CentOS, and numerous other Linux distributions. Any organization running Exim 4.97–4.99.2 with GnuTLS on an internet-facing mail server is at risk.
The attack surface is enormous: every SMTP relay, mail gateway, and email forwarding service running affected versions is a potential target. Unlike client-side vulnerabilities, mail servers are designed to accept connections from untrusted sources — there's no user interaction required, no phishing email to click. The attacker just needs to open an SMTP connection and send the malicious BDAT sequence.
The fact that AI was used to develop the exploit lowers the barrier for other threat actors to replicate the attack. Once the technique is public, expect scanning and exploitation attempts to begin within days.
How to Protect Yourself
- Upgrade to Exim 4.99.3 immediately: This is the only complete fix. Run
apt update && apt upgrade exim4(Debian/Ubuntu) or your distribution's equivalent - Verify your Exim build: Run
exim -bV | grep gnutlsto confirm whether your build uses GnuTLS. If it uses OpenSSL, you're not affected by this specific vulnerability - Restrict SMTP access: As a temporary mitigation, limit which IPs can connect to your SMTP port (25/587) while you patch
- Monitor for exploitation: Watch Exim logs for abnormal BDAT session terminations and unexpected process crashes — heap corruption may cause Exim child processes to segfault before code execution succeeds
- Consider switching to OpenSSL builds: If your distribution offers an OpenSSL-built Exim package, it's not vulnerable to this specific flaw (though you should still patch for other reasons)
The Sable Angle
Mail servers are the crown jewels of infrastructure — they handle authentication emails, password resets, and internal communications. An unauthenticated RCE on an MTA isn't just a server compromise; it's a potential gateway to every account and service that depends on email.
At Sable, our infrastructure penetration testing includes MTA-specific attack simulations — BDAT fuzzing, SMTP command injection, and TLS downgrade attacks. We test whether your mail infrastructure can withstand the kind of protocol-level attacks that automated scanners miss entirely.
This is also the second AI-assisted zero-day disclosed this week. The threat model has changed: attackers don't need deep expertise in heap exploitation when AI can help them get there. Your defenses need to assume the adversary has machine-speed exploit development.