AI

In a FinOptiCorp scenario, attackers exploit LLM "FinBot" via prompt injection and OWASP flaws for data leaks, RCE, and model theft—Trend Micro's Vision One™ AI Security provides AI Scanner, Guard, and layered defenses to secure generative AI against backdoor risks.

OpenAI's October 2025 report reveals bans on ChatGPT accounts linked to PRC-affiliated groups like UNKDROPPITCH, who leveraged AI to debug malware (GOVERSHELL, HealthKick), craft targeted phishing, and build surveillance tools—disrupting 40+ networks; models block direct threats but highlight efficiency risks in cyber ops.

LayerX uncovers CometJacking, exploiting Perplexity’s AI browser via malicious URLs to steal Gmail/Calendar data—tricks AI into Base64-encoded exfiltration, bypassing safeguards. Urgent call for AI security-by-design amid rising agentic threats.

A malicious update to the postmark-mcp server injects a hidden BCC to exfiltrate sensitive emails from thousands of organizations. Koi’s risk engine uncovered the attack, highlighting risks in AI-driven MCP tools. Remove version 1.0.16+ and audit MCP servers now.

Microsoft exposes AI-driven phishing campaign targeting US organizations: attackers use AI to craft verbose, business-jargon code in SVG attachments disguised as PDFs, hiding credential-stealing payloads behind invisible dashboards and evading antivirus detection.

India's universities endure 7,095 weekly cyberattacks—higher than global averages—due to hybrid models, limited resources, and connected campuses. Check Point report highlights RATs, malware risks; experts urge prevention-first security, AI monitoring, and investment to protect data and research.

Security scan reveals 150+ popular apps (millions of downloads) with Firebase test mode flaws allowing unauthenticated access to payments, PII, chats, passwords, and GitHub/AWS tokens in Realtime DB, Storage, Firestore, and Remote Config. Learn impacts, OpenFirebase tool, and fixes for ~80% of mobile apps.

Threat actors use in-memory PE loaders to download and run malicious executables (e.g., RATs) via Windows APIs like VirtualAlloc and LoadLibraryA, evading file-based EDR like Microsoft Defender/Sophos. Learn the technique's steps, red team success, and need for memory/behavioral defenses.

Average breakout time drops to 18 minutes (June-August 2025, per ReliaQuest), fueled by automation and Oyster malware's abuse of rundll32.exe for DLL loading via scheduled tasks. Learn about Gamarue USB attacks, AI-driven malvertising, and defenses like behavioral monitoring.

SentinelLABS uncovers MalTerminal, an early LLM-enabled malware using OpenAI's GPT-4 to dynamically create ransomware or reverse shells at runtime, evading detection and marking a new era in AI-driven cyber threats.