Oussama Rahali

In the past week, AI code editor Cursor disclosed five HIGH-severity issues.  Here’s the summary: 1️⃣ CVE-2025-59944: RCE Via Sensitive File Overwrite Bypass ➖ Case-sensitive checks protecting Cursor’s sensitive files could be bypassed on case-insensitive filesystems. With prompt injection, an attacker could overwrite those files to achieve RCE. 2️⃣ CVE-2025-61590: RCE Via Prompt Injection ➖ VS Code/Cursor Workspaces automatically keep a `.code-workspace` file that stores project settings. If an attacker hijacks the chat context (e.g., via a compromised MCP server), they can prompt the agent to write into the workspace settings, editing protections to allow for RCE. 3️⃣ CVE-2025-61591: RCE Via MCP OAuth2  ➖ Using OAuth with an untrusted MCP server allows a malicious server to return crafted instructions during the auth/interaction flow, leading to command injection, and ultimately RCE under the user’s privileges. 4️⃣ CVE-2025-61592: RCE Via Permissive CLI Config ➖ Cursor CLI auto-loads a project-local config when you run the tool inside that repository. If the repo is malicious and that config permits shell execution, prompt-delivered rules can steer the agent to run arbitrary code. 5️⃣ CVE-2025-61593: RCE Via Sensitive File Overwrite in CLI ➖ The Cursor CLI Agent’s protection of sensitive files could be subverted via prompt injection, allowing file modification and RCE. These CVEs are the culmination of cyber hygiene and novel AI vulnerabilities: ▪️ untrusted MCP servers serving malicious injections in returned fields, ▪️ case-folding & normalization bugs, ▪️ weak separation between chat context and filesystem/command execution, ▪️ overly permissive RBAC As I said in my last post: >> AI risk management is more than securing AI models. << AI systems inherit the risk of the ICT environments they run in, as well as bringing their own vulnerabilities. #AISecurity #Cursor #MCP #AI #CyberSecurity

1 004

43 commentaires

🚨 CVE‑2025‑47812: Critical RCE Vulnerability in Wing FTP Server A high‑severity vulnerability, CVE‑2025‑47812, has been identified in Wing FTP Server (versions prior to 7.4.4). With a CVSS score of 10.0, this flaw stems from improper handling of null bytes in the username parameter, allowing unauthenticated attackers to inject arbitrary Lua code into session files, leading to root/SYSTEM-level remote code execution. 💡 Why it matters: Exploiting this vulnerability could result in: * Full system compromise via remote code execution with SYSTEM or root privileges. * Active exploitation observed in the wild, including privilege escalation, user creation, and remote payload delivery.   🛡️ Recommended actions: * Upgrade Wing FTP Server: Immediately update to version 7.4.4 or later, which resolves the vulnerable session handling logic. * Block inbound HTTP/S traffic: Apply firewall rules to block inbound TCP ports 80 and 443, reducing exposure until the server is patched.We got your back! Use these scripts from the Vicarius research team: 🔍 Detection script: https://lnkd.in/d3jefWkD 🩹 Remediation script: https://lnkd.in/dvqfGAtd Remediate CVE-2025‑47812 with #vRx by Vicarius: https://lnkd.in/dXedBin6

56

FortiGuard Labs has released an Intrusion Prevention Service (IPS) update version 33.049, to detect and block attack attempts targeting CVE-2025-53770 a critical zero-day vulnerability in Microsoft SharePoint. It has been actively exploited in the wild since at least July 18, 2025. This vulnerability affects on-premises SharePoint Servers and currently has no available patch. Microsoft has confirmed that attackers are targeting this flaw, which appears to be a variant of the previously known CVE-2025-49706.

27

2 commentaires

🚀 CipherMQ 1.3.1 Released! What's New: 🔐 Dynamic CN-based Key Distribution: Replaced static client_id with Common Name (CN) from mTLS certificate (client.crt) for secure key management. Public keys stored in SQLite using CN, encrypted with AES-GCM. 🔑 Hybrid Encryption: Switched to x25519 for session key encryption, keeping AES-GCM-256 for messages to ensure Perfect Forward Secrecy. 💻 Code Enhancements: Updated receiver for CN extraction, sender for CN-based key retrieval, and added storage for SQLite key management. https://lnkd.in/dB-D9C-H What is CipherMQ in a nutshell: CipherMQ is a secure message broker in Rust that uses hybrid encryption (RSA + AES-GCM) to send encrypted messages between Python clients. It guarantees zero message loss and exactly-once delivery through acknowledgment mechanisms and message deduplication. News: From now on, you will hear about CipherMQ at the following address (it's under construction): https://ciphermq.com/ Try #CipherMQ 1.3.1 today and enjoy the most possible security in your data stack. Ehsan Masoumi

44

2 commentaires

Last week the The New York Times reported that the Chinese government-backed threat actors known as Salt Typhoon who are primarily targeting “government, transportation, lodging, and military infrastructure networks” and may have stolen data from “almost every American.” The attacks have targeted at least 80 other countries – essentially every country that is a geopolitical competitor to China is being targeted for widespread surveillance. One primary goal of these attacks is to compromise telecommunications infrastructure – a system known as “Signaling System No. 7” (SS7) which has a variety of different names in different countries. When attackers compromise SS7 they are able to track the location of mobile devices even if users of those devices use encrypted tools within those targeted phones – essentially if you have a cell signal, you can be tracked. The attacks also have the ability to monitor unencrypted SMS messages, voicemails, and also “listen in” on unencrypted phone calls. There are reports that President Trump and members of his administration were targeted – and it seems likely that President Biden and his administration would have been targeted too. If you are a VIP politician or CEO of a major industry, it’s very possible that the targeting would have focused some collection on your devices and those of your family and close friends. Over the coming weeks (hopefully not months), Cybersecurity and Infrastructure Security Agency has agreed to release their 2022 report on the insecurity in the telecommunications industry which facilitated these attacks. Hopefully this report will include more details about *how* these attacks happened. Were all of these done via niche attacks on legacy hardware? Was there ever any phishing infrastructure? What did the C2 domains/IPs look like which sent the stolen data to Chinese servers? Our team at Silent Push has been sitting on some research about Salt Typhoon for a while after sharing it with our clients earlier this year. We recently decided to share all of the ~45 domains that we’ve tracked associated with these Chinese attacks. The dozens of domains we’re releasing were used by Salt Typhoon and a separate Chinese APT group known as UNC4841 which is currently tracked separately from Salt Typhoon but has similar victim targeting and we’ve proven in our research release that they shared commonalities when registering domains – very specific fingerprints. What this means is that 2+ Chinese APT groups that are state sponsored and heavily targeting ~every large country were for ~5 years using a domain registration fingerprint which made it possible to identify their domains before they were weaponized and used in attacks. Our team found domains dating back from 2020 until early in 2025, nearly the full timeline of the Salt Typhoon attacks which likely started in 2019. .... (More in comments + research links) ⤵️

38

3 commentaires