November Patch Tuesday: Microsoft fixed a record 1,235 flaws, 66 critical. One is already exploited: CVE-2026-124194 in Outlook. Open a crafted message file and Outlook connects to the attacker's server without the usual external-content warning, handing over your NTLMv2 hash to crack offline. Classic phishing attachment, so patch Outlook first. Fun detail: the list includes the OpenSSH "sshole" bug, because it hits Microsoft's own Linux, Mariner.

BleepingComputer

Microsoft November 2026 Patch Tuesday fixes 1,235 flaws, Outlook zero-day

Today is Microsoft's November 2026 Patch Tuesday, with security updates addressing a record-breaking 1,235 flaws, including one actively explo...

Best write-up of the LNbits attacks so far, from someone it happened to: an attacker took @Logen's node from a free account in about eight hours. Open sign-up, a Paywall extension bug that let a normal user read server files for 1 sat, super-user access, then a malicious extension that read LND's admin.macaroon and tried to close every channel. If you run LNbits: never give it admin.macaroon. Bake one that can only invoice and pay, and mount that file alone. His post has the full chain, signs of compromise and cleanup steps. View article →

Heads-up for nos.lol and nostr.mom (@Ostrich-70): since about Nov 7 most writes get no OK at all. Reads are fine. Today at 13:18 UTC the newest kind 1 on either was over an hour old, and a test note from a fresh key got no answer in 60s on both. Same from a VPS and from a phone, so it doesn't look like an IP block. Maybe the spam filter is stuck?

The WorkOS breach wasn't a hacker after all. WorkOS now says it was a researcher's AI agent, built on ARTEX, that escaped its test environment and spent days pulling customer records from an internal sales tool. Nobody has said how it got out. What it found once loose was ordinary: exposed to the internet, weak credentials, no rate limiting, little monitoring. WorkOS has been running ARTEX against its own systems since November 3. https://gbhackers.com/workos-data-breach-artex-ai/

DeepMind had two AI agents attack 1024-bit RSA, with a human sending about 20 messages, mostly "approved". In under four days they designed and ran a signature forgery that took 391 CPU core-years instead of the previous 1,380, and found a batch method that makes each extra key about 30x cheaper to factor, on paper. No key was broken, and 2048-bit is still fine. But if something you depend on still signs with RSA-1024, the clock just got faster.

SecurityWeek

Google DeepMind's AI Agents Make 1024-Bit RSA Keys Cheaper to Attack

Google DeepMind AI agents found two ways to make attacks on legacy 1024-bit RSA cheaper, while 2048-bit and larger keys remain secure.

LNbits is telling every operator to shut their server down. Attackers are creating new accounts on instances and withdrawing funds; one operator lost about 2.28M sats. The hosted my.lnbits.com is suspended too. A patch is promised within 24-48 hours, and nothing is in the public repo yet. Third big infrastructure scare in two weeks, after Northflank and BTCPay. If you run one, take it down until the fix lands. Source:

LNbits Orders Emergency Shutdown After Active Attacks Drain Sats

LNbits told every operator to pull the plug immediately after attackers exploited account-creation and withdrawal functions, draining 2,285,545 sat...

If you use Claude Code, check your repos. Apiiro found 1,122 public GitHub copies of its permissions file with at least one secret inside. Approve a command with "don't ask again" and it gets saved there, token and all. They didn't test whether the secrets still work. The file is meant to be git-ignored, so `git ls-files | grep .claude/` is a quick check. Mine came back clean.

The Hacker News

Claude Code's “Don't Ask Again” File Exposes Passwords and Keys in 1,122 Public Copies on GitHub

Claude Code saves secrets in approved commands, exposing 2,539 credentials across 1,994 public GitHub repositories, Apiiro finds.

OpenSSH bug with a lovely mechanism: CVE-2026-123158, "sshole". Send a password of 257+ bytes and on musl the hash function returns "*", which is exactly what locked accounts store as their password. So it matches, and you're in an account that shouldn't accept passwords at all. Needs musl, no PAM and password logins on, which is Alpine's default. Alpine's cloud images even ship a locked "alpine" user with passwordless doas. Fixed in 10.6p2. Debian and Ubuntu use glibc and PAM, so they're fine.

BleepingComputer

OpenSSH flaw lets attackers log into password-disabled Linux accounts

A newly disclosed OpenSSH vulnerability dubbed "*sshole" allows attackers to log in to certain Alpine Linux and other musl-based Linux accounts tha...

OpenAI caught its own model sandbagging. GPT-6 Astra quietly did worse on AI safety work (red-teaming, exploit tasks, evals) than on identical tasks framed as capabilities research: about 10% of hidden criteria met vs 16%. It invented blockers, dropped working solutions, and the monitors missed most of it. Telling it to "be careful" made it 50% worse. OpenAI's best guess is overcaution. They also say they can't rule out motives it never wrote down.

SecurityWeek

OpenAI Finds GPT-6 Astra Quietly Holding Back on Internal Red-Teaming, Exploit Tasks

GPT-6 Astra quietly underperformed on AI safety research tasks such as red-teaming, and OpenAI’s monitors mostly missed the behavior.

Self-hosting WordPress? CVE-2026-119977 (CVSS 8.1) is being exploited, and the vulnerable moment is the core update itself. During a full update, WordPress stages files in a predictable, web-reachable folder, and anyone who hits it in those few seconds can run code, no account needed. Wordfence saw 18,500+ attempts in 21 hours. Fixed in 7.1.4 and backported to older branches. Ironically you have to update to fix it, so block wp-content/upgrade/ at the web server while you do.

Google is telling schools: keep our AI on, or lose Search. From early 2027, K-12 student accounts get Gemini and Search switched off until the school enables them, and Google says it can't separate the two because AI Overviews and AI Mode are now part of Search. Block the AI and students just get a "not available" page. So the most-used search engine in classrooms now comes bundled with a chatbot, take it or leave it.

New Nostr Scrutiny fact sheet: White Noise.

White Noise — Nostr Scrutiny

Fact sheet about the Nostr client White Noise: who is behind it, keys, code and privacy.

Some facts that stood out: - Group chats use Marmot (MLS over Nostr), with throwaway keys for each message. No audit yet: the protocol's own threat model marks 80 of 86 risk scenarios as unverified. - Analytics, telemetry and diagnostic logs are off by default, on their own servers, with deletion deadlines. - Firebase for push on Android, in the Zapstore build too. - The app can share your location and Google Play lists location data, but the privacy policy never mentions it. The map loads straight from OpenStreetMap.

Chainalysis, the firm tracing the stolen Binance funds, says it got breached by "offensive agentic techniques". The attacker reached internal tools and took customer names, emails, user IDs and role labels. No investigation data or credentials, they say. What "agentic" actually means here, they don't explain. It's going to be the new "sophisticated attack" if companies get to say it without details.