Chris Choi

🚨Invariant Labs has uncovered "Toxic Flows," a novel class of vulnerabilities that expose agentic AI systems to prompt injection and MCP-based exploits. To help developers stay ahead, we’re introducing the Toxic Flow Analysis (TFA) framework and the MCP-scan tool, the industry’s first approach to systematically detect and mitigate these emerging AI security threats before they happen. As AI agents become more dynamic and autonomous, traditional security tools fall short. TFA represents a paradigm shift, helping developers analyze how sensitive data flows through an AI system, and where things can go dangerously wrong. Dive deeper into TFA and try out MCP-scan in our latest blog: https://bit.ly/4lTnRPK

66

🔓 We built local backdoors for Signal, 1Password, and Slack using a “data” file their integrity checks ignored. CVE-2025-55305: override common JavaScript builtins in v8_context_snapshot.bin. When apps use these builtins during normal operation, malicious code executes. The vulnerability affects nearly every Chromium-based application because code integrity checks don't consider these snapshot files "executable"—but they contain JavaScript that runs with the privileges of their V8 isolate. Even apps with Electron's integrity protection fuses on were vulnerable to this framework-level bypass. https://lnkd.in/g4QHijVs

328

9 Comments

Broken API authorization is still exposing sensitive data - so Intruder built a FREE tool to find it.⚡ Three years after the Optus breach - where millions of customer records were stolen via an unauthenticated API - Intruder’s security team is still finding the same flaws, even in S&P 500 companies. These vulnerabilities are so easy to exploit you could teach someone with no technical background to do it in a day. 🤯 That’s why Intruder built Autoswagger - a FREE, open‑source tool that finds unauthenticated API endpoints leaking sensitive data like credentials or PII. Get Autoswagger FREE on GitHub and see the real vulnerabilities Intruder’s security team uncovered when they put it to the test. Get Autoswagger: https://lnkd.in/e8cA3Ktr 🔥 Read the research ->> https://lnkd.in/eKZz_ybJ

29

Heads-up, BFSI: Chrome’s EKU change on June 15, 2026 is an enforcement, not a relaxation. Public-trust TLS leaf certificates must include EKU and assert only id-kp-serverAuth (no mixed purposes). The aim is to reduce Web-PKI risk by driving dedicated TLS hierarchies for public TLS. 📖 Full blog: https://lnkd.in/gkstmmNq Why this matters for finance if you’re using public CA certs beyond browsers—it can impact: 🔐 Mutual TLS (mTLS) 🔗 API authentication 💳 Server-to-server banking communications Path forward: ✔ Transition these use cases to Private PKI trust models (on-prem or managed) ✔ Use finance-sector PKI for inter-bank trust (e.g., ASC X9 Financial PKI) ✔ Build crypto-agility with automated Certificate Lifecycle Management (CLM) At eMudhra, we’re helping banks, central banks, and payment processors modernize with resilient, compliant PKI—spanning Private PKI and CLM (CertiNext)—so critical flows stay secure and uninterrupted. #BFSI #TLS #Chrome #PrivatePKI #CLM #DigitalTrust #Cybersecurity #mTLS #APIsecurity Arvind Srinivasan Kaushik Srinivasan, CFA Biju Varghese Virender Aggarwal Scott Rea, CISSP Carmine Auletta Spencer Frye Manikantan Dakshinamoorthy Rajesh (Raj) Raman Gita Kao PRASANNA NARASIMHA Gishnu S Nambiar Nutan K Arju Krishnan Sai Prasad Bikash Singh Janarthanan S Srinivas Prasad Sudesh Kumar Rajat Gurnani (Raj)

84

Your image, your rules. 🤳 We've partnered with the Cyber Security Cooperative Research Centre and the University of Chicago to develop a new technique to stop unauthorised AI systems learning from photos, artwork and other image-based content. The method subtly alters content to make it unreadable to AI models while remaining unchanged to the human eye. The breakthrough could help artists, organisations and social media users protect their work and personal data from being used to train AI systems or create deepfakes. While the method is currently applicable to images, there are plans to expand it to text, music, and videos. Read the story: https://lnkd.in/gi5HxXFE 📷 | dtatiana via Adobe Stock

67

5 Comments

🚨#Security Alert - Largest #npm compromise in history: We've been tracking this for npm attack for the past 24 hours, and it's still unfolding. A malicious actor compromised the npm registry by publishing trojanized versions of 18 widely-used packages, after obtaining developers’ tokens in a phishing attack, reported by Aikido Security. The total download count of these packages amounts to ~2 billion downloads weekly and several extremely popular packages were compromised including debug, chalk, ansi-styles and more. Additionally, we’ve detected the malicious payload in another package, [email protected] + our monitoring infrastructure also detected that DuckDB has also been compromised. Learn more and get the latest updates in this developing attack: https://jfrog.co/42kAtrh

116

2 Comments

OPSWAT’s Unit 515 discovered a critical vulnerability (CVE-2025-24293) in Ruby on Rails Active Storage that could allow remote code execution in certain configurations.  The research exposed a bypass of a previously patched CVE and contributed directly to the Rails security team’s latest fix.  Get the full breakdown: https://hubs.la/Q03JhWQg0

39