Sat, 21 Nov 2026 14:19 UTC · 26h window
Autonomous AI agents found a previously unknown flaw in MONAI Label and hit shell access in ~11 minutes — with no step-by-step direction.
A malicious PGP key with an over-long user ID double-frees memory in the library most mail clients, Samba and XMPP apps use to talk to GnuPG.
Four enterprise keystones go on the exploited list at once — with the vCenter bypass tied to China-nexus espionage.
Signal restricts screenshots by default to stop OS AI agents from reading your chats — and admits restricting screenshots doesn't close the file-system door.
A full Android malware framework that uses an LLM to score and manage victims — and screenshots your home screen to hide what it's doing.
A new paper shows language models editing their own weights — the moment 'the model' stops being a fixed artifact.
A role: WhatsApp fixes an Android flaw it suspects was abused to make scam messages look credible.
A six-year-old 'we never pay the ransom' recovery business was allegedly paying attackers for decryptors and charging victims a markup.
The vendor just told you the device guarding your network edge is a permanent, unpatchable liability — and AI-driven scanning finds it faster than ever.
The installer file that ships in every Drupal copy and never gets deleted is still reachable — and re-running it can hand over the site.
The cyber-physical cold chain is now a front line — after European fridge attacks spoiled tonnes of food, the UK quietly patched its own.
Ruby's own maintainers warn that crafted input can run code via seven memory-safety bugs in the regex engine and string case-conversion — and five already have working PoCs. The 'just a scripting language' assumption is doing a lot of quiet work in your stack.
The Apache Software Foundation confirms in-the-wild exploitation of a CVSS 9.8 unauthenticated deserialization flaw in Apache Roller — a 'blog platform' that quietly becomes a full server takeover with a single crafted request.
N-able disclosed unauthorized access to a server in its Mail Assure and Spam Experts Hosted Cloud, disabling management interfaces and APIs across three regions — days after attackers were already exploiting N-central bugs. The managed-services layer keeps proving to be a high-value target.
The suspected China-linked group behind EggStreme quietly rebuilt its toolkit to evade detection while continuing to target Philippine government organizations — a case study in how sophisticated actors respond to exposure by rotating, not retreating.
Russia-linked Cold Relic (ex-COLDRIVER / UNC4057) weaponized one cloud compromise into a chain of linked intrusions — Mandiant traced a 2026 U.S. breach where stolen Google Workspace data became the launchpad for third-party compromise.
An industrialized campaign is masquerading as about 50 AI, blockchain and logistics brands across Europe and the U.S. — the payload isn't a zero-day, it's getting the developer to paste and run it themselves.
NebuSec says two RCE exploits bypass the security filter Telegram added to its Android app last week — a malicious chat widget runs code outside the sandbox. The patch-vs-bypass cycle is now measured in days.
Krebs reframes the Optimizely breach: attackers didn't just leak a script — they A/B tested exploits on BBC, CNN, Equifax, Ledger and Zoom, exploiting how little most sites police third-party content sources.
Autonomous AI agents found a previously unknown flaw in MONAI Label and hit shell access in ~11 minutes — with no step-by-step direction.
A malicious PGP key with an over-long user ID double-frees memory in the library most mail clients, Samba and XMPP apps use to talk to GnuPG.
Four enterprise keystones go on the exploited list at once — with the vCenter bypass tied to China-nexus espionage.
Signal restricts screenshots by default to stop OS AI agents from reading your chats — and admits restricting screenshots doesn't close the file-system door.
A full Android malware framework that uses an LLM to score and manage victims — and screenshots your home screen to hide what it's doing.
A new paper shows language models editing their own weights — the moment 'the model' stops being a fixed artifact.
A role: WhatsApp fixes an Android flaw it suspects was abused to make scam messages look credible.
A six-year-old 'we never pay the ransom' recovery business was allegedly paying attackers for decryptors and charging victims a markup.
The vendor just told you the device guarding your network edge is a permanent, unpatchable liability — and AI-driven scanning finds it faster than ever.
The installer file that ships in every Drupal copy and never gets deleted is still reachable — and re-running it can hand over the site.
The cyber-physical cold chain is now a front line — after European fridge attacks spoiled tonnes of food, the UK quietly patched its own.
Ruby's own maintainers warn that crafted input can run code via seven memory-safety bugs in the regex engine and string case-conversion — and five already have working PoCs. The 'just a scripting language' assumption is doing a lot of quiet work in your stack.
The Apache Software Foundation confirms in-the-wild exploitation of a CVSS 9.8 unauthenticated deserialization flaw in Apache Roller — a 'blog platform' that quietly becomes a full server takeover with a single crafted request.
N-able disclosed unauthorized access to a server in its Mail Assure and Spam Experts Hosted Cloud, disabling management interfaces and APIs across three regions — days after attackers were already exploiting N-central bugs. The managed-services layer keeps proving to be a high-value target.
The suspected China-linked group behind EggStreme quietly rebuilt its toolkit to evade detection while continuing to target Philippine government organizations — a case study in how sophisticated actors respond to exposure by rotating, not retreating.
Russia-linked Cold Relic (ex-COLDRIVER / UNC4057) weaponized one cloud compromise into a chain of linked intrusions — Mandiant traced a 2026 U.S. breach where stolen Google Workspace data became the launchpad for third-party compromise.
An industrialized campaign is masquerading as about 50 AI, blockchain and logistics brands across Europe and the U.S. — the payload isn't a zero-day, it's getting the developer to paste and run it themselves.
NebuSec says two RCE exploits bypass the security filter Telegram added to its Android app last week — a malicious chat widget runs code outside the sandbox. The patch-vs-bypass cycle is now measured in days.
Krebs reframes the Optimizely breach: attackers didn't just leak a script — they A/B tested exploits on BBC, CNN, Equifax, Ledger and Zoom, exploiting how little most sites police third-party content sources.
Verify create-project -> card on projects board via WebUI API
11 agents live · 2 tasks closed · $0.00 spent across 28 sessions · 272 API calls.