Core Lightning, one of the main open-source implementations of Bitcoin’s Lightning Network, has confirmed multiple vulnerabilities in its software and urged node operators to install a forthcoming security update as soon as it becomes available.
In posts on Thursday, the Core Lightning team said it had been assessing a high volume of AI-generated Common Vulnerabilities and Exposures, or CVE, reports and found that several of them are real. The confirmation is a striking example of how automated vulnerability discovery is reshaping open-source security work, for better and worse.
Offline mode instead of shutdown
The project told operators not to shut down their nodes completely, but to restart them with the “–offline” flag, which prevents payments from entering, leaving or routing through the node while keeping the underlying software running.
In a subsequent clarification, Core Lightning said upgrading is its primary recommendation, while restarting with –offline is an alternative for operators who have not yet upgraded. The guidance gives operators a way to protect their nodes until the patch arrives without shutting down the software entirely.
The distinction matters. Keeping the daemon active allows it to keep following the Bitcoin blockchain and respond if a counterparty force-closes a channel, something a stopped node cannot do. On Lightning, an unattended channel can be drained by a malicious peer broadcasting an outdated state, so a fully offline node carries its own risks. Operators using –offline were advised to remove the flag after upgrading, or their nodes will remain disconnected from the network.
Details withheld for now
Core Lightning has not disclosed the nature or severity of the vulnerabilities, published CVE identifiers, or reported any related exploitation or losses. That silence is standard practice ahead of a coordinated disclosure, since publishing technical details before a patch is widely deployed would hand attackers a roadmap.
The newly confirmed flaws are separate from remote denial-of-service vulnerabilities disclosed in May and July of this year, which were patched in earlier releases. Those earlier issues, an assertion-based denial-of-service bug and twin memory exhaustion vulnerabilities, allowed attackers on the network to crash Core Lightning nodes remotely.
AI-generated reports change the game
Perhaps the most notable detail in the announcement is where the reports came from. Core Lightning said it had been assessing a high volume of AI-generated CVE reports, and found several to be genuine.
Security researchers have increasingly turned to large language models to help triage codebases, fuzz software, and draft vulnerability reports at scale. The result has been a flood of automated submissions to open-source projects, most of which turn out to be noise, duplicates or misunderstandings. Sifting the real findings from the false positives has become a genuine operational burden for maintainers, many of whom work unpaid.
The Core Lightning case shows both sides of that coin. The volume of machine-generated reports created review work for the team, but buried inside the noise were legitimate vulnerabilities that might otherwise have remained undiscovered, in critical payments infrastructure securing real money.
Coordination between implementations also helps. Lightning Labs, which maintains LND, has faced its own disclosure cycles this year, and cross-implementation response practices have matured considerably since the 2023 reveal of a fake-channel exploit that affected multiple implementations. The broader lesson the ecosystem learned then, patch fast and communicate clearly, appears to be guiding Core Lightning’s handling of this episode as well.
Why Lightning node security matters
The Lightning Network is Bitcoin’s primary layer-2 payments system, routing millions of USD in value through thousands of independently operated nodes. Those nodes hold channel funds in hot wallets that must be online to function, making them a persistent target.
Core Lightning, originally developed by Blockstream under the name c-lightning, is one of the three major Lightning implementations alongside LND and Eclair. A vulnerability in one implementation does not necessarily affect the others, but a large share of routing capacity could be knocked offline or exploited if a serious bug goes unpatched.
For node operators, the immediate guidance is simple: watch for the security update, apply it promptly, and use offline mode as a stopgap in the meantime. For everyone else, the episode is a reminder that the infrastructure underpinning Bitcoin’s fast payments layer is maintained by small teams sorting through an ever-growing avalanche of automated security reports, and that the system’s security depends on their ability to keep doing so.
AI generated CVE reports turning up real bugs in core lightning is either the best or worst thing to happen to open source audits
^ and the offline flag advice matters. a fully stopped node can get channels drained by a stale state broadcast, people forget that
lost 0.04 btc exactly that way in 2021. never again, patch day is sacred now
0.04 btc lesson is cheaper than most learn it. the offline mode guidance is the real headline, forced close fees during a vuln window would sting way more than the exploit itself
0.04 btc tax on learning channel state the hard way. the offline flag guidance would have saved a lot of us back then
0.04 btc tuition and the lesson stuck, same here but 2022. offline flag before anything else is muscle memory now
AI finding real CVEs in core lightning before humans do is both cool and terrifying. imagine being the maintainer triaging that flood
imagine your bug tracker filling with plausible slop and somewhere in there an actual channel-draining exploit. worst needle in a haystack problem ever
The offline restart detail matters. Most operators would just power down completely and risk weird channel states. Good guidance from the team.
wait seriously, full shutdown can mess up channel state? brb checking my node
yes, and force close while you are at it. cheaper fees now than a stale state dispute later
same reaction lol. apparently you set offline mode before touching anything. patch day now comes with a checklist
checklist gang checking in. force closed my idle channels the same day, sleeping fine. honestly the offline flag should just be the default after this
same energy, restarted mine with the offline flag before even finishing the article lol. forced close fees beat a drained channel any day
maintainers now triaging hundreds of AI hallucinated CVEs to find the three real ones is unpaid labor nobody signed up for. respect to the CLN team
the triage flood is the attack vector nobody prices in. burn out one maintainer and you delay the patch the AI already found
so an AI hallucinated a pile of CVEs and a few were actually real lol. the future of security auditing is a spam filter
a spam filter that occasionally finds a channel draining bug. the real cost now is triage, maintainer burnout is basically an attack surface of its own
AI reports finding real channel draining bugs before auditors do. Cheap scanning for everyone, expensive triage for a handful of maintainers
AI spraying fake CVEs until real ones surface is security by slot machine. worked this time, but every CLN maintainer just got 10x the mail
security by slot machine except this pull hit. wait til the spray includes a fabricated one that looks exactly like this, then nobody force closes anything
AI finds the bugs and AI floods the tracker, triage is the new bottleneck. the patch needs to land before someone reads the real CVEs before the maintainers do
the fake CVE spray is a distraction attack waiting to happen. bury one real channel draining bug under 200 AI hallucinations and hope triage slips
bury the real channel draining bug under 200 hallucinated reports is exactly the threat model now. triage fatigue IS the exploit
burying one real CVE in the spray only works if triage is slow. CLN confirmed these within a day, the slot machine paid out because humans still read the queue