Strong piece from SourceCodeRed https://t.co/96chDhn7ll If traditional SCA struggles to spot malicious packages, what sets JFrog’s malicious-package detection apart? Our approach combines in-house scanners, curated public databases focused specifically on malicious packages

Our research team uncovered five backdoored npm crypto packages that operated for about six months, injecting SSH keys and leaking cryptographic entropy. Read the full analysis and learn how they persisted, what they targeted, and how to protect your supply chain:

🚨 New Zero-Day #Vulnerability: Our team found a vulnerability in oatpp-mcp (CVE-2025-6515) that lets attackers hijack sessions. But the real story is the type of attack. Our new blog post explores this "session hijacking" vector for MCP servers - get the full breakdown:

Our research team found 3 malicious MCP servers on PyPI with 1.6K downloads, executing harmful payloads! mcp-runcmd-server (XRAY-734538) mcp-runcommand-server (XRAY-734540) mcp-runcommand-server2 (XRAY-734539) These packages promise command execution but spawns a reverse shell

🚨 #CyberSecurityAwarnessMonth Mini-Series: "#MCP Attacks" Our team at JFrog uncovered #CVE-2025-6514, a critical RCE vulnerability (CVSS 9.6) in mcp-remote. Let's talk about it & the exploitation scenarios👇🏼 mcp-remote acts as a proxy, enabling apps that only support local

🚨 #CyberSecurityAwarenessMonth Mini-Series: "MCP Attacks" uncovers... Exploiting MCP Servers with Prompt Injection RCE: How do MCP Prompt Injection Attacks directly impact the #SoftwareSupplyChain? ☝🏼An #MCP prompt injection happens when a user with access to the host’s LLM

One more visual to complete the set 👇

🚨 #CyberSecurityAwarnessMonth Mini-Series: "MCP Attacks" - What to Watch Out For & How to Stay Safe 🛡️ Scroll through to see what attack techniques to expect when building #MCP capabilities in your org.👇🏼 👉🏽 Here's the breakdown: Trends quickly become attackers’ favorite

📣 #CyberSecurityAwareness Mini-Series: "Hunting Malicious ML Models" Part 5, THE FINALE asks... 🫨 Can "Safe" ML Models be Weaponized?⚡️ ✨ Let’s start by defining what a “Safe” Models is: Safe models were created to reduce this malicious code execution risk we discussed in

📣 #CyberSecurityAwareness Mini-Series: "Hunting Malicious ML Models" Part 4 is all about... 🏃🏻 Library Mitigations and How They Can be Bypassed 🏃🏻 Scanners cannot take full responsibility for protecting against malicious models. This has pushed #ML Python libraries to add

Another Redis vulnerability, CVE-2025-46817, has been classified as Critical by the NVD. This vulnerability is an integer overflow issue that, similar to CVE-2025-49844, requires an authenticated attacker capable of executing a specially crafted LUA script to exploit it.

📣 #CyberSecurityAwareness Mini-Series: "Hunting Malicious ML Models" Part 3 poses the question... How do scanners work and why do they generate False Positives (FPs) + False Negatives (FNs)? 🤔 ❎ False Positives (FPs) ❎ Looking at the numbers, other scanners on Hugging Face

It is recommended to upgrade Redis to a patched version immediately (e.g., 8.2.2 and above), especially for servers that are exposed to the internet. To protect JFrog customers, Xray's DB and the JFrog Catalog have been fully updated to guarantee detection and prevention of this

This flaw allows an attacker to send a specially crafted malicious Lua script (a feature supported by default in Redis), which escapes from the Lua sandbox and possibly to achieve remote code execution on the Redis host. The attack works because the malicious Lua script tricks

Exploitation of the RediShell vulnerability requires the attacker to be authenticated. While this narrows the risk to either Redis servers deployed with NO authentication enabled or scenarios involving malicious internal/authenticated users, the potential of remote code execution

The new 10 CVSS score vulnerability CVE-2025-49844 that was published for the widely used, in-memory database Redis, dubbed as “RediShell”, is a use-after-free memory corruption bug that has been present in the Redis source code for approximately 13 years!

📣 #CyberSecurityAwareness Mini-Series: "Hunting Malicious ML Models" Part 2 talks about the 3 most common model threats 👇🏼 🔹 PyTorch-Malcode (1,263 files detected) #PyTorch models’ default save and load functionalities rely on Pickle serialization and deserialization.

Too many #security tools are creating more problems than they solve. This #CyberSecurityAwarenessMonth, JFrog Field CISO, Paul Davis, breaks down why a fragmented defense can leads to higher costs, slower remediation, and a weaker security posture. Learn how to move from silos

🔐 October marks #CybersecurityAwarenessMonth! We kicked it off yesterday with our first miniseries: “Hunting Malicious ML Models – Insights from Analyzing 1.5M Models.” And all month long, we'll be sharing new research, blogs, & webinars as your go-to resources for uncovering

Our security team uncovered a malicious PyPI package called SoopSocks, which disguises itself as a SOCKS5 proxy but behaves like a backdoor. Our research revealed that it installs persistence via Windows services and scheduled tasks, modifies firewall rules, silently executes