Anza, the developer of Solana’s Agave validator client, on Sunday evening told mainnet validators to take action on CVE-2026-98432, a Linux kernel bug that can let a remote, unauthenticated attacker run code on a server. The alert, labeled “URGENT,” came before Ubuntu had a fix: Anza’s requirements page says the team builds and runs its software on Ubuntu 24.04. Anthropic has said one of its unreleased research models found the bug and built a working exploit.
Macro strategist Jim Bianco, in a Bits + Bips episode recorded the day before Anthropic’s Oct. 27 post, said that now “we’ve got industrial scale AI going through all of this” in search of flaws in “tens of billions of lines” of code written “by hand one at a time.”
Get Unchained’s crypto news in your inbox with the free Unchained Daily newsletter.
Still Waiting on Ubuntu
Anza’s post shared one from Tim Garcia of the Solana Foundation, who wrote that the flaw “is patched in the linux kernel but not yet patched on ubuntu,” and pointed operators running Ubuntu to steps they could take in the meantime.
As of Monday afternoon, Ubuntu’s security tracker listed the standard kernels for both 24.04 and 22.04 as vulnerable, with fixes in progress and none released. The standard 24.04 kernel’s fix was in progress as version 6.8.0-149.149; some newer 24.04 kernels, including the linux-hwe-6.17 and linux-hwe-7.0 lines, had none. Ubuntu rates the bug high priority. The 9.8 severity score on the National Vulnerability Database record comes from the organization that issued the CVE; NIST has not yet scored it.
AWS shipped fixed kernels for Amazon Linux 2 and Amazon Linux 2023 and said a kernel update and reboot “is the only complete remediation for this issue.” As interim steps it listed turning off multipath TCP, which it said “does not resolve the issue,” and a rule that drops IPv4 packets carrying IP options. AWS said it knew of no exploitation in the wild. Red Hat, which rates the flaw Critical, said its own packet filter “reduces exposure to this vulnerability but cannot fully prevent it.” It also cautioned that filtering “may disrupt necessary workloads” and told customers to update once fixes are out.
AWS said the bug can be triggered with specially crafted packets sent to an IPv4 TCP port that is reachable from the internet. Anza’s requirements page says a validator needs a public IPv4 address.
The alert comes less than two weeks after Solana activated Alpenglow on mainnet, its new consensus engine.
Found by an AI Model
In an Oct. 27 research post, Anthropic said Claude wrote an exploit that lets an attacker with no credentials run commands as root on a Linux server with a multipath TCP listener, just by sending it network packets. Anthropic said its proof of concept needed 48 minutes to expose a kernel memory address, then five more minutes to start running shell commands.
AI tools have turned up validator bugs before: in July, the Ethereum Foundation said its AI agents had found a real validator bug.
Related Listen: Anthropic’s Mythos Has More Capabilities Than the U.S. Government. Is That Right?
