Rapid Response Round Up: Two LLM-flavored bugs

|
Updated

On the heels of our very bad October, I wanted to take a second here to discuss a couple of vulnerabilities that surfaced in November that piqued my interest. We usually provide a quick review of the month’s most noteworthy bugs during runZero Hour, affectionately known as the Rapid Response Rundown, but this month we ran out of time (thanks to our amazing guest, Thai Duang of Calif). So, rather than let them go unloved, here's a written version of this month’s Rapid Response Rundown. I suspect these two bugs are harbingers of a theme that we’re likely to see evolve over the next few weeks and months: kind of boring plumbing that nobody looks at, but when an LLM-assisted vuln-hunter does look, unexpected vulnerabilities fall out.

OpenSSH and the very long password (CVE-2026-123158) #

Covered in our Rapid Response writeup, CVE-2026-123158, has its own cutely named site, sshhole.com. This vulnerability lets you log into some OpenSSH servers by pasting a very long password (257 to 1024 characters) at the prompt for an account that has no password. Oopsy.

The bug lives in the gap between two components that each look correct. Locked accounts have a * in /etc/shadow, which should never match anything. But musl's crypt() returns "*" on failure instead of NULL, and its SHA-512 crypt fails on passwords over 256 bytes. So, a too-long password "hashes" to *, matches the locked account, and you're in. It needs musl, no PAM, password authentication on (the default), and at least one other account with a real hash on the box. OpenSSH 10.6p2 fixes it.

There's a little bit of finger-pointing going on at who was actually responsible for this failure (OpenSSH says the root cause is musl, while musl says it's OpenSSH's bad choices). The POSIX standard says crypt() should return NULL on failure, so it looks to me like musl is more at fault, but of course, the fix ultimately came from OpenSSH.

The practical exposure is the long tail. IoT and embedded gear is more likely to ship musl (rather than glibc) because it's smaller, so OpenWrt and Yocto devices are directly implicated, along with some unknowable number of other resource-constrained platforms. Alpine Linux also needs patches. The official Alpine cloud images on AWS and Azure ship a default user with passwordless doas, so this bypass quickly becomes root.

Without a patch, set PasswordAuthentication no, which you should be doing here in 2026 anyway. Alas, lots of off-the-shelf gear won't expose this easily, so pester your vendor for a firmware update to include this fix.

Incidentally, Özgün Kültekin found this by building sshd every way imaginable and throwing about 17,600 weird logins at it, with an assist from an LLM swarm.

Apache HTTP/2 heap corruption (CVE-2026-119618) #

CVE-2026-119618 is a race condition in Apache httpd's mod_http2. A body-less HTTP/2 request with a Content-Type but no Content-Length makes a worker thread write a synthesized Content-Length: 0 into a shared header table. If that table is full, the write allocates from the connection's allocator, which has no mutex, while another thread is using it. Two threads end up owning the same memory, and the heap gets corrupted. No authentication is required. It affects httpd 2.4.55 through 2.4.70, and is fixed in 2.4.71.

Interestingly, this issue was discovered and reported twice inside 8 days. Anthropic's OSS Scanner reported it on October 26 (as ANT-2026-MAPKT6MZ), and Anthony Weems and Simon Scannell of Google Cloud's vulnerability research team reported it around November 3. Credit lines now list scanners with bug IDs next to named humans, I guess. While I'm not thrilled with "crediting" an LLM with a "discovery," I expect more of that to come in CVEs going forward.

The severity of this issue seems to be disputed between RedHat and Wiz, which is kind of interesting, and there's no public report of known exploitation yet.

If you can't upgrade yet, remove h2 and h2c from Protocols and fall back to HTTP/1.1. I'm of the opinion that today, most sites outside the mega-CDNs aren't reaping the performance gains that HTTP/2 provides, and earlier HTTP/2 bugs like Rapid Reset are implying that HTTP/2 implementations are kind of rickety compared to the battle-tested HTTP/1.1 deployments.

Finally, for a deep dive into how mod_http2 handles concurrency and some excellent post-facto vuln detective work, there's a superb write-up from googlesky for your reading pleasure.

Two bugs with LLM pedigrees #

Both of these bugs sit in interactions between components, and not bad coding practices, internal type confusions, or other common root-causes of vulnerabilities: libc semantics against an application's assumptions (for OpenSSH and musl), and thread safety in a shared pool allocator (for Apache mod_http2).

Machines are getting good at finding this kind of issue, and humans and machines are converging on the same bugs fast enough that OpenSSH skipped their usual embargo period. So, patch windows are shrinking, as we've been warning for a while.

The common thread is the attack surface you forgot about. Turn off password authentication, turn off HTTP/2 if you don't need it, and know where your embedded OpenSSH servers are hiding. Knowing what's on your network before the next bug lands is what runZero is for, so pop over to runzero.com/try for a fully functional free trial.

You can watch the most recent episode of runZero Hour right here, and if you want to catch the next episode live, subscribe to runZero Hour to join us next month—hopefully with slightly better time management (but no less interesting of a guest!).

Written by todb

Tod Beardsley is VP of Security Research at Dragos, where he "kicks assets and fakes frames." He joined Dragos following its acquisition of runZero, where he worked after leaving public service. Prior to 2025, he was the Section Chief for the Vulnerability Response section for CSD/VM/VRC at CISA, the Cybersecurity and Infrastructure Security Agency, part of the US government, and a seasonal Travis County Election Judge in Texas. He's also a founder and CNA point of contact for AHA!. Tod spends much of his time involved in vulnerability research and coordinated vulnerability disclosure (CVD). He has over 30 years of hands-on security experience, stretching from in-band telephony switching to modern ICS/OT implementations. He has held IT ops, security, software engineering, and management positions in large organizations such as the US Government, Rapid7, 3Com, Dell, and Westinghouse, as both an offensive and defensive practitioner. Tod is a CVE Board member, has authored several research papers, and is an internationally-tolerated horror fiction expert.

More about todb
Subscribe Now

Get the latest news and expert insights delivered in your inbox.

Welcome to the club! Your subscription to our newsletter is successful.

Explore more runZero

Product
runZero 5.1 is here: Secure AI workflows, enhanced integrations, and expanded autonomous discovery
runZero 5.1 takes on the heavy lifting across five key areas, enabling you to unmask and remediate exposures with less friction and more speed.
Podcasts
Know Your Adversary with HD Moore
runZero CEO HD Moore breaks down the myth of air-gapped networks, the impact of AI on security, and why asset connectivity is everything.
runZero Perspective
BOD 26-04: A new era of prioritized remediation
A complete breakdown of CISA's BOD 26-04 directive. Learn how the shift to SSVC, risk-based KEV prioritization, and 3-day remediation impacts your...
runZero Perspective
Dawn of the apex agentic adversary
When agentic AI can weaponize exploits in seconds, visibility is everything. Stop the predator with runZero’s exposure management for the AI-attack...
Webcasts
runZero Hour, Ep. 36: Agentic AI – because finding bugs manually is so 2025
HD Moore and Thai Duong joined the runZero Hour to discuss AI bug hunting, vulnerability disclosure, PHP2Shell mechanics, and the October 2026 PHP...
Podcasts
Risky Business: Tod Beardsley talks post-quantum cryptography
VP of security research Tod Beardsley joins Patrick Gray to chat about fear around the looming post-quantum cryptography crisis.
Webcasts
runZero Hour, Ep. 35: Dragos and the Research Cinematic Universe
Recap the runZero Hour crossover with Dragos, NetRise, and runZero. Discover insights on the PHPocalypse, OT/IT security research, and...
Podcasts
Risky Biz: HD Moore talks OT security, frontier fearmongering and more
Patrick Gray chats with industry legend HD Moore about how his company runZero wound up being used heavily to discover OT devices in enterprise...

See Results in Minutes

See & secure your total attack surface. Even the unknowns & unmanageable.