The Hacker News

🛑 LiteLLM CVE-2026-42208 exploited in ~36 hours. A pre-auth SQL injection exposed credential tables with LLM and cloud keys—turning a simple flaw into account-level risk. No PoC needed; advisory and schema were enough. 🔗 Read details → https://thehackernews.com/2026/04/litellm-cve-2026-42208-sql-injection.html

🔥 GitHub RCE via single git push! CVE-2026-3854: Unsanitized push options let attackers run commands on backend servers, bypassing sandboxing (cross-tenant risk). 🔗 Learn how header injection led to full compromise → https://thehackernews.com/2026/04/researchers-discover-critical-github.html Patched within hours.

⚠️ A Brazilian cybercrime group is back, targeting Minecraft players with a fake mod. LofyStealer runs directly in memory, stealing passwords, tokens, and banking data across major browsers after a single install. 🔗 Read more → https://thehackernews.com/2026/04/brazilian-lofygang-resurfaces-after.html

Google flags indirect prompt injection as a primary attack vector for AI agents. Attackers embed hidden commands in websites and documents—seen by AI, not humans—redirecting execution away from user intent. Detections rose 32% in recent scans. The risk sits in what models consume. 🔗 How web content is being weaponized against AI systems → https://thehackernews.com/2026/04/weekly-recap-fast16-malware-xchat.html#:~:text=Turning%20the%20Web%20Into%20a%20Trap%20for%20LLMs

🛑 VECT 2.0 ransomware can’t restore what it destroys. Files over 131 KB are permanently destroyed, not encrypted—because required nonces are discarded during execution. No attacker can recover the data, even after payment. 🔗 Read → https://thehackernews.com/2026/04/vect-20-ransomware-irreversibly.html

🚨 9 days ago, a compromised OAuth token resulted in a breach at Vercel. This is the reality facing security teams: 🥷 An infostealer hits a vendor in your SaaS ecosystem, the attacker works backwards from stolen access, and an old consent becomes the front door. 🚪 If you haven't made OAuth grant reviews a regular part of your on-going security program, you should. Here's a step-by-step guide from Nudge Security to help you → https://thn.news/oauth-checklist

Security teams treat data as trusted once it crosses domains. That’s the ⚠️ flaw. 53% still rely on manual processes, even as attacks hit 137 times a week and 🤖 AI drives response speeds on both sides. The gap isn’t identity. It’s how data moves. 🔗 Why data movement is the weakest layer in Zero Trust → https://thehackernews.com/2026/04/why-secure-data-movement-is-zero-trust.html

⚠️ An unpatched critical flaw in Hugging Face’s LeRobot enables remote code execution (CVSS 9.3). Untrusted pickle over unauthenticated gRPC (no TLS) lets attackers take over servers, steal keys and models, and impact connected robots. 🔗 Details → https://thehackernews.com/2026/04/critical-cve-2026-25874-leaves-hugging.html

AI has erased the patch window. Exploit discovery now takes minutes, not weeks—leaving organizations exposed before fixes exist. Security shifts from prevention to real-time containment. 🔗 How AI-driven exploits are forcing an assume-breach model → https://thehackernews.com/2026/04/after-mythos-new-playbooks-for-zero.html

A Chinese national linked to Silk Typhoon has been extradited to the U.S. over alleged COVID-19 research cyberattacks. Prosecutors say Xu Zewei exploited zero-days to breach vaccine research systems under direction of China’s MSS. 🔗 Details → https://thehackernews.com/2026/04/chinese-silk-typhoon-hacker-extradited.html

A user ran malware and no alerts fired... As Ryan Boerner, Founder and CEO at Keep Aware, shows, the browser is now the real OS, but tools below it miss in-session actions where AI and scripts act as the user, breaking identity trust. 🔗 Learn how browser-based attacks bypass modern security stacks → https://thehackernews.com/expert-insights/2026/04/work-moved-into-browser-security-didnt.html

😳 Entra ID Agent ID Administrator role flaw enabled service principal takeover. Users could take over non-agent service principals, add credentials, and escalate privileges before Microsoft’s April 9, 2026 patch. 🔗 See how the attack worked → https://thehackernews.com/2026/04/microsoft-patches-entra-id-role-flaw.html