Gotenberg's fix for the default deny-list bypass deletes the default deny-list, hides the real check behind a flag, and ships the flag off. The advisory asked them to reconsider the defaults. They did.
https://t.co/hi41YUM25I
Kava Lend's interest math divides by what's left in a market, with one check: skip it if the market is empty. Attacker left two units in a dead BUSD market, division ran, 53.9M KAVA walked. Fifth time this decade.
https://t.co/IgkWsbLCb7
We wrote up the protobufjs bug from the issue Cristian Staicu of @EndorLabs filed: a type name that was JavaScript.
https://t.co/dPsuBSmzrf
Corrections welcome.
The @synacktiv relay demo is in our case file on CVE-2026-26128, which argues the advisory title names the wrong component.
https://t.co/wouQ0X7T6t
If we have the relay path wrong, tell us.
We built this case file on the @bishopfox UniFi chain write-up: the auth gate read the raw URI and nginx routed the normalized one.
https://t.co/7B9byLLC1o
Corrections welcome.
The @qualys advisory on CVE-2026-46333 is the source for this case file, including that the fix was proposed on the kernel list in 2020.
https://t.co/RxkFenRQvz
If we misread the advisory, tell us and we'll correct it.
We took the @cyera_io n8n disclosure down to the patch: three handlers, one assertion each.
https://t.co/A0hNzoBg14
It is one of four files behind our read of what n8n keeps doing:
https://t.co/4vaMf0xOyn
The @X41Sec LiteLLM sandbox bypass is the spine of this case file: thirty-four regex patterns, and a fix that deleted the file.
https://t.co/Yuqi84LLyS
If we have your timeline or your bypass wrong, tell us.
We wrote up both Langflow bugs @Horizon3ai found, down to the patch.
The validate endpoint compiles and runs: https://t.co/7SHtLC2M9I
flow_id authorized, data ran: https://t.co/qljCw1GSQh
Corrections welcome.
Thirteen of our case files lean on @watchtowrcyber write-ups and detection artifacts. Each one reads the patch down to the diff. Start with cPanel's session file:
https://t.co/v5YC2BqDUC
If we got your work wrong anywhere, tell us and we'll correct it on the page.
221 advisories since April, and six habits account for 129 of them. Vendors aren't shipping bugs, they're shipping reruns with a press office.
https://t.co/bBUNse1yHC
vm2 shipped four independent defenses in three days for a leak Node already fixed. The maintainer's own doc says they assume one will fail. That's the job now.
https://t.co/KVtfLlKc8H
Flowise shipped the same bug three times: a list with no criterion, a scrubber with no type, a parser that is the runtime. Each patch changes what sits inside the name. The name is the bug.
https://t.co/NWhaRKzSnZ
Ivanti's whole security model is documents. The integrity checker is a document. The CVSS rating is a document. The end-of-life page is a document. The attacker reads all of them.
https://t.co/HCCMJN0zLO
Fortinet ships three products where "authentication" means the caller types their name twice and the server nods. Then they issue IOCs built on the same field the attacker chose.
https://t.co/P0DmzZ4Vtm
Cisco shipped three CVSS 9+ bugs this year in three different protocols, all the same shape: the parser answers the network before the auth check runs. Workaround is turn the feature off. The parser is the feature.
https://t.co/GimXs8I5tP
BerriAI keeps shipping CVEs their own repo already knew how to prevent. Co-founder imported the sandboxed Jinja2 class in 2024. Four later developers imported the unsandboxed one. Nobody read one directory over.
https://t.co/Dqj16o1Ly3
n8n has shipped the same bug four times in a year. Every patch adds a check at the handler. The helper that would kill the class sits one floor up, untouched. It is a fucking policy.
https://t.co/4vaMf0xOyn
Linux ships seven of these a year. Two files, both correct, nobody paid to read them together. The fix is two lines. The wait is years.
https://t.co/3Bncnl0oCQ
Google named a parameter elidable and the function never read it. Do that seven times across seven products and it's not a bug, it's the house style.
https://t.co/9F4Z99EOL8