LINKED LIST [txt mode] ▸ @pubkey ▸ Blog
home explore | log in

Blog

by @pubkey · last added 2026-11-21 · 58 items
* GnuPG advisory for GPGME 2.3.1 (CVE-2026-129975): an OpenPGP key with a crafted very long user ID triggers a double free in every prior version, usable to exfiltrate data from apps and RCE can't be ruled out; disclosure moved up from Nov 25 because GitLab had to patch early for Ruby2Shell
lists.gnupg.org · node›
* phpAfterShell (CVE-2026-126714, CVSS 9.2): PHP's file session handler can be interrupted mid-write during unauthenticated upload-progress updates, splicing the prefix of one valid session onto the tail of another so the next load deserializes attacker-controlled data, which depthfirst chained with a PHP-native memory corruption bug into RCE
depthfirst.com · node›
* Wiz details nginx2root: two critical unauthenticated HTTP/2 RCEs in NGINX, a flow-control integer underflow dating back to 1.9.5 (2015) that Wiz exploited on a server just serving static files, and a proxying use-after-free; found independently by five teams and fixed in F5's Nov 18 release
wiz.io · node›
* Optimizely's Web Experimentation incident report: a flaw in the account signup flow let an attacker impersonate admin users without credentials and inject card- and crypto-skimming JavaScript into customer projects; first malicious write 07:51 UTC Nov 16, detected 13:34, closed 16:10, and sites with a strong CSP were protected
pages.optimizely.com · node›
* Microsoft Security Research phishes a GitHub device-code approval in a lab and shows what one click hands an attacker on an AI coding agent: remotely steering a live Copilot CLI session to run commands, planting false agent memories, reading cloud-synced sessions, and pulling private repos
microsoft.com · node›
* Hello Again (ACM CCS 2026): rolling back fTPM lockout state through SPI flash replay-protected counters brute-forces a 4-digit Windows Hello PIN in 40 hours instead of 69 days, via a Winbond HMAC key leak under voltage glitching or AMD skipping increment checks
doi.org · node›
* Microsoft FORGE Lab's AI-assisted Linux kernel hunt: 26,502 raw findings boiled down to 621 PoC-backed cases and 59 upstream reports, yielding 33 CVEs with eight confirmed root exploits; the bugs averaged 8.1 years old and turning scans into merged fixes took months of human work
microsoft.com · node›
* Flow/Neos critical (CVSS 9.8) fix: uploads were processed before the action's security policy ran, and extension blocklist gaps let an unauthenticated request drop scripts or an .htaccess into Web/_Resources/Persistent
neos.io · node›
* PostgreSQL's Nov 12 update fixes 52 CVEs across 14–18 plus 19 RC2: dozens of 8.8 type-confusion and use-after-free bugs that let a SQL user execute code, pg_dump/pg_restore and reindexdb injection paths, and logical replication writing to subscriber catalogs; PG 14 is now EOL
postgresql.org · node›
* Git's Nov 10 releases fix six bugs back to v2.27: a 32-bit ce_namelen wrap turns a validated .gitz/z…z path into .git so recursive clones can plant a hook-running symlink, plus a merge-buffer heap overflow forges can hit via git merge-tree
openwall.com · node›
* OpenSSL's Nov 10 advisory: an empty incremental AAD update silently resets AES-GCM-SIV's AAD so tags stop covering earlier fragments, plus QUIC leaking uninitialized stack bytes into the peer address; five Low CVEs, releases Dec 1
openwall.com · node›
* hyperglass looking glasses drop user targets into router commands with a bare str.format(); AS-path quotes, IPv6 scope-ID text after % and IOS newlines all become pre-auth RCE — 36 hosts across 20 ASNs, 11 root
hyperpwn.xyz · node›
* Home Assistant: refresh-token secrets and per-token JWT keys sit in cleartext in .storage/auth, long-lived tokens never expire, and changing your password revokes none of them — so one leaked backup or config read becomes durable access to the house
openwall.com · node›
* UnrealIRCd CVE-2026-114713: an escaping bug in the IRCv3 label tag lets any user inject newlines into server-to-server traffic and run commands with server authority; 6.2.7.1 and a no-restart hot-patch fix it
openwall.com · node›
* sshole: on musl sshd without PAM, accounts with shadow hash '*' accept any password of 257–1024 chars because musl crypt() returns '*' for oversize input — OpenSSH 10.6p2 / PasswordAuthentication no
sshhole.com · node›
* InfiniteWP Client NOSSL: signing key is md5(panel_pubkey || rand(0,2³¹)) — recovered live in 292s from one captured request; forge forever once enrolled
openwall.com · node›
* Check Point: ChatGPT Lockdown Mode bypassed via nested iframe window.open — silent chat exfil and phishing (fixed)
research.checkpoint.com · node›
* bREACHYing the Perimeter: JFrog chains five CVEs in Pollen Robotics Reachy Mini — unauthenticated FastAPI on *:8000 plus passwordless sudo and a CORS/CSRF browser path to persistent root and motor control
jfrog.com · node›
* CWP2Shell: VulnCheck's ten-CVE chain in Control Web Panel — a login redirect with no exit() lets an unauthenticated module include run, then a root loopback user API escalates to uid 0; fixed in CWP 1.29+
vulncheck.com · node›
* Cloudflare's follow-up on eMudhra's eight absurdly weak CA keys (one was literally the number 4): what exploitation they actually observed, how fast Pingora and internal systems were patched, and how slowly visitors' trust stores are catching up
blog.cloudflare.com · node›
* Atredis takes Steam apart: two remote code execution bugs and four local privilege escalations in Valve's client, with full PoCs released after disclosure through HackerOne
atredis.com · node›
* ZDI on CVE-2026-35273: a PeopleSoft Integration Gateway SSRF lets attackers loop requests back through 127.0.0.1 to pass validateClient and reach RCE endpoints, the chain ShinyHunters has been mass-exploiting
zerodayinitiative.com · node›
* Kaspersky GReAT: an unauthenticated firmware-upgrade endpoint on Intelbras IP cameras lets anyone on the local Wi-Fi knock them offline, and repeating the request every minute keeps them down indefinitely; fixed in firmware R.260804
securelist.com · node›
* Anthropic's Frontier Red Team details five attack chains an unreleased internal model found and exploited in browsers and operating systems, reaching remote code execution with little or no user interaction, all reported under 90-day coordinated disclosure
anthropic.com · node›
* JFrog on Brotli2Shell (CVE-2026-111497): the out-of-bounds write is in Google's Brotli decoder, not Node.js, and they reach it with no HTTP at all, getting instruction-pointer control via a crafted JPEG XL file in FFmpeg plus native crashes in .NET and a 534-byte infinite loop in pyarrow
research.jfrog.com · node›
* Vercel's $1M Sandbox escape challenge: 1,285 reports in two weeks, $242K paid, no customer data reached, and the most severe finding was an upstream Linux kernel zero-day found from a standard sandbox on day three
vercel.com · node›
* Microsoft on the poisoned LiteLLM 1.107.0 wheels: payloads ran from Python startup hooks (.pth and site config), so merely starting Python or pip in an affected env could fire the credential stealer, and uninstalling the package isn't recovery
microsoft.com · node›
* gnark's fake-GLV scalar multiplication never constrained the prover's result point to lie on the curve, so in-circuit ECDSA over P-256, P-384 and the STARK curve (including EIP-7951 P-256 verification) accepted forged signatures for any key; fixed in v0.16.4
certora.com · node›
* A malicious remote MCP server can run PowerShell on Windows and WSL through an https:// OAuth URL in OpenCode and MCP Inspector, because open@10 wraps the target in double quotes and $(...) survives every scheme check
corgea.com · node›
* PHP2Shell: Wiz chains decade-old memory-safety flaws in the PHP engine itself into unauthenticated RCE and Heartbleed-style memory leaks back to PHP 5.4, patched Oct 21 under two CVEs up to CVSS 9.2
wiz.io · node›
* FortiSIEM CVE-2026-106173: a gRPC service on tcp/27918 binds every interface with no client-cert check, and SendQueryResult builds a file path from attacker-supplied fields, giving unauthenticated arbitrary file write to RCE
horizon3.ai · node›
* WPSplinter: a Broadcom WPS parser bug in the SDK since at least 2010 gives pre-auth RCE over WiFi, with only an incomplete 2018 patch and no CVE, so it's a forever day on eCos routers
quentinkaiser.be · node›
* BlockStrap (CVE-2026-103849): Magento's StyleSmuggler hotfix signed trusted directives, but child-template data still gets signed and rescanned at root, so a smuggled {{block}} runs PHP unauthenticated
sansec.io · node›
* Xoscar CVE-2026-49950: every actor-pool message is cloudpickle-loaded before its type is checked, with no auth on the TCP transport, so any reachable pool is unauthenticated RCE, still unpatched in v0.11.2
vulncheck.com · node›
* Big Fish: a signed 16-bit shm_nattch counter from the original BSD sources wraps after 259 forks, and a race plus a -1 OOB write turns it into OpenBSD root past KARL, SMAP and SMEP
calif.io · node›
* Hekate 0.35/0.36: S=t support left unsalted helper leaves fixed, so a valid proof could confirm a witness guess offline; 0.37 adds one support cell
oumuamua.dev · node›
* XBOW's $250K Chrome full chain: a V8 compiler bug, an animated-WebP decoder bug, and a Crashpad overflow took a tab to unsandboxed host code execution in six days
xbow.com · node›
* GoBalance: truncating Tor-format Ed25519 keys dropped the nonce PRF material, so one public descriptor recovers the onion master scalar
slcyber.io · node›
* Google DevSite RCE: a Flask session cookie carried the OAuth secret it was signed with, so forging a pickle session took two GET requests
simonkoeck.com · node›
* NetScaler log-to-root teardown: a command injected into a log file, later run as root by the appliance's own maintenance script (CVE-2026-88771/88772)
threatunpacked.com · node›
* Project Zero: how big vendors ship emergency security fixes faster than their normal pipeline (feature flags, filtering, alternate channels, hotpatching)
projectzero.google · node›
* OpenSSH 10.6: LZ77 compression disabled over cross-channel plaintext recovery, GSSAPI and sftp path fixes
openssh.com · node›
* Smashing the token limit: CSS-injection exfil of long tokens via overlapping fragments
portswigger.net · node›
* Too Small to Hide: single-trace key recovery from ML-KEM keygen
eprint.iacr.org · node›
* Recovering SNOVA secret keys from biased vinegar sampling
eprint.iacr.org · node›
* Incomplete FO ciphertext compare in ML-KEM: IND-CCA2 break to key recovery
eprint.iacr.org · node›
* OpenSSL Security Advisory (29 Sep 2026): DTLS heap disclosure, EC/SM2 timing leaks
openssl-library.org · node›
* MAMBA-Frost: lattice KEM from Learning With Quantization
eprint.iacr.org · node›
* Breaking the Gómez-Torrecillas–Lobillo–Navarro cryptosystem with LLL
eprint.iacr.org · node›
* Forging 1024-bit RSA signatures in nearly SNFS time
eprint.iacr.org · node›
page 1/2 next >