Microsoft shows AI agents backdoored by editing under 10 neurons
The Lazarus method needs checkpoint write access, not retraining, and Microsoft has not seen attackers use it.
1
Stories per day · 30 days
15 RSS
Relatedsupply-chain25wordpress10data-poisoning6rank-math6aomei5poisoning4unc70174android3botnet3claude-code3cursor3linux3plugin3shinyroad3updraftcentral3updraftplus3videostrong3backupper2
203 stories, newest first · one card per story
Stories tagged #backdoor
CERT-In Flags AI-Driven Attacks Using Backdoored ARTEX Image
CERT-In discloses active attacks by the Crafty Falcon gang targeting India's BFSI sector and crypto firms with AI tools like ARTEX, while a modified Docker image for the tool grants root shell on every unauthenticated…
GARD cuts CLIP visual backdoor success below 1%
The post-hoc defense covers six visual backdoor attacks on two CLIP backbones and needs no extra inference compute.
SURA detects federated-learning backdoors with 10 clean images
It reuses one small clean proxy across repeated loss-ascent probes and restores batch-norm state only at readout.
LoRA subspace projection suppresses backdoors but does not erase them
Few-shot reactivation restored 91.7% to 100% attack success in one epoch for every weight-editing defense tested.
Anubis ransomware deploys Lindnor backdoor in US attack
Anubis ransomware operators broke into a US organization deploying custom Lindnor backdoor and 13 security-killing tools including vulnerable drivers.
Reference answers poison backdoors in skill optimization
Poisoned reference answers induce persistent trigger-conditioned backdoors in SkillOpt.
Hackers Deploy CWPdoor Backdoor via Trojanized CWP Server
A new Linux malware called CWPdoor is distributed through trojanized versions of cwpsrv in the Control Web Panel web server, affecting over 340,000 exposed instances.
Goddess of Victory: NIKKE PC launcher tampered in supply-chain attack
Hackers modified NIKKE launcher scripts to inject remote-access malware stealing credentials and crypto wallets; update issued November 4.
1
Three defenses harden face recognition
Three tricks detect poisoned identities at training and filter inputs at inference.
China-Aligned Hackers Hijack Sogou Input Method to Deploy DarkVigil Backdoor
Attackers abuse DNS hijacking in Sogou Input Method to deliver DarkVigil backdoor to Chinese-speaking users across Asia.
3
Hackers hijack cPGuard autoupdate to deploy backdoor
Attackers compromised OpsShield's CDN to distribute a backdoor and data-stealing script through cPGuard updates, affecting nearly 3,000 servers.
cPGuard Update Channel Hijacked to Install Backdoors
Attacker hijacked cPGuard update system to distribute backdoor in versions 5.88.12-5.88.19 that collected password hashes and other files from Linux servers.
Videostrong backdoored 1.3 million Android TV boxes
Google discovered Videostrong backdoor in 1.3 million Android TV boxes that turned them into residential proxy botnet used by criminals.
1
Attackers exploit Node-RED CVE-2026-58115 to deploy Linux bot and backdoors
VulnCheck detects in-the-wild exploitation of CVE-2026-58115 deploying a Linux bot and webshells.
Hackers Compromise cPGuard Update Channel to Install Backdoor
Attacker accessed cPGuard distribution infrastructure to distribute malicious script installing backdoor and collecting files from Linux servers.
3
Videostrong backdoor powers 1.3M Android TV box botnet
Coalition disrupts XCPE backdoor in Videostrong devices infecting 1.3M Android boxes across 200 countries.
Google disrupts 1.3 million-device Android botnet with Videostrong backdoor
Google and partners took down a botnet of 1.3 million Android TV boxes containing a persistent backdoor sold by device maker Videostrong.
Malicious Claude Code extension reached up to 5,000 Cursor users
A marketplace backdoor stayed live about 11 hours, but a design flaw stopped it on most devices.
1
Trojanized Claude Code extension for Cursor hit developers
Malicious version 2.1.313 stole data and, in a minority of cases, could give persistent privileged access.
1
Backdoors claimed in official Claude Code VS Code and Cursor extensions
The linked post body did not load, so only the title states the backdoor claim.
Pretraining poison can backdoor Qwen3 CLI agents for RCE
Poisoning 0.1% of pretraining data led the 4B agent to about 92% attack success, and success stayed positive at 0.0005%.
Sirius backdoor variant delivered by Korean FedEx-themed PDF emails
Trellix says a weaponized PDF launches fileless loaders that inject Sirius into suspended Windows processes after the victim approves a prompt.
RePAIR repairs backdoored semantic segmentation models without retraining
The post-hoc, label-free method infers victim-target pairs from unlabeled images and corrects suspicious pixels.