SafeDep

MemTensor packages on npm and PyPI were compromised today by a Go worm called sckit. The attacker didn't push malicious code directly. They got into MemTensor's own CI pipelines, intercepted the publish tokens before the real release job ran, then used those tokens to ship the backdoor themselves. No install hook involved. --ignore-scripts does nothing here. Full breakdown with IOCs → https://lnkd.in/dvrU7vXF

Catch PMG at IndiaFOSS and take part in some fun activities.

🛠 Arsenal Announcement: BSides Mumbai 2026 Dependency scanning usually catches a malicious package after it is already in the lockfile, already on a developer's machine and often already executed via an install hook. By then the interesting part has happened. Sudhanshu Dasgupta, Software Engineer at SafeDep, brings "PMG (Package Manager Guard): Block Malicious Packages at Install Time" to the Arsenal track. What PMG does Intercepts the install itself, so a known-malicious package is blocked at the point of entry rather than flagged after the fact. About Sudhanshu He works on infrastructure for open source supply chain security at SafeDep and contributes to tooling that helps developers detect and block malicious packages before they reach production. He is a core maintainer of Meshery, an open-source CNCF sandbox project for cloud-native infrastructure management, and speaks regularly on open source, supply chain security, cloud native technologies and community building. He actively mentors new contributors in the open source ecosystem. Who this is for DevSecOps and platform engineers, AppSec teams, open source maintainers and anyone who owns dependency policy. 🗓️ 3 October 2026 📍 Mumbai, India 🔗 Full lineup and tickets: https://bsidesmumbai.in Where does your org actually block a bad dependency today, at install, at CI or at review? Would be good to hear what's working. ♻️ Repost so your platform team sees it. . . . . #BSidesMumbai #SupplyChainSecurity #DevSecOps #OpenSource #AppSec #CloudNative #CNCF #SoftwareSupplyChain #Cybersecurity #InfoSec

PMG v0.29.0 is out, and it's a big one on reach. Three things worth calling out: - Windows support. PMG now runs on Windows, tested end to end in CI. System install, doctor checks, and the full protection layer are all there, so teams on Windows dev machines get the same install-time malware blocking that mac and linux users already had. - New package managers. The aub family (aube, aubr, aubx) is now intercepted alongside npm, pip, yarn, bun, uv and the rest. More of your install surface covered with the same zero-config setup. - Sandbox any program. The sandbox used to wrap package installs. With the new pmg sandbox exec, you can run any program inside PMG's OS-native sandbox with restricted file and network access, not just installs. Useful any time you want to run something you don't fully trust. PMG is free and open source. Full changelog and setup: github.com/safedep/pmg

A package on npm called pino-ulid looks like a ULID generator. It is a remote access trojan. The real ulid package is a tiny, boring utility that npm downloads about 8 million times a week. That is what makes the name work. "pino-ulid" reads like harmless plumbing, so it slides into a project without a second look. The moment you install it: 1. A postinstall hook (dist/node/utils.js) runs during npm install, before you ever import the package. 2. It sets up Windows persistence through a scheduled task and registry key named pkg-agent, then spawns a hidden agent (dist/node/payload.js). 3. The agent connects to a hardcoded WebSocket C2. The operator writes code to disk and runs it. Full remote code execution. pino-ulid@2.12.3 C2: ws://95.216.232.162:8010 (XOR and base64 obfuscated) 

When you move from VS Code to Cursor, what happens to your extensions? https://lnkd.in/dmJ2aMyR

A scanned image doesn't mean a safe Kubernetes Pod. Scanning runs at build time. But dev workspaces, build Pods and CI jobs keep installing npm and pip at runtime, long after the scan. A package published an hour ago can land straight in a running container. PMG now checks those installs inside the container and blocks the malicious ones before they run. No account, no Kubernetes API access and no extra RBAC. Full Kubernetes setup, including how each workload gets a stable identity: https://lnkd.in/dgGNnKTJ

The real-time face-swap tool a lot of people are messing around with just got hijacked. A GitHub repository with over 90,000 stars, a real-time face-swap tool was compromised through a maintainer account takeover, 2FA and all. The maintainer's account was pointed to one dependency at a fake repo. Install it, and it quietly swaps any crypto address you copy for the attacker's. No release shipped it, but for about 9 hours anyone who installed from source got hit. Full breakdown (in comments)

Introducing SafeDep Threat Intel. Ask a SOC team how they find out a dependency in their network went malicious, and the honest answer is usually `too late`. The package gets published, pulled into a build, and executed. The advisory lands days later. By then the window that mattered has closed. We spend our time on the other side of that window, analyzing and documenting real npm, PyPI, Go and Maven campaigns as they happen. Today we're making that same intelligence available to SOC and cyber defense teams directly, as SafeDep Threat Intel. Two ways to use it: - A malicious package feed API. A continuous stream of suspicious and confirmed malicious packages, campaigns, and IOCs across npm, PyPI, Go, Maven, and more. Query it or push it into the tools your team already runs. - A Threat Intel dashboard in the SafeDep console. Every verdict, the reasoning behind it and the campaigns that connect related packages. ->https://lnkd.in/diVYjgya

We've been building something new at SafeDep, for organizations and developers who like to stay a step ahead. Announcing soon...