Hi, I'm Vivian, a cybersecurity and AI Product Manager trying to keep up with an industry that moves faster than I can whisk up my morning matcha. Every week brings a new wave of vulnerabilities, AI security mishaps, and breaches that keep us on our toes, so I take some time to share the most interesting news instead of letting it all blur together. Let's dive into whats been keeping us up at night recently in cybersecurity and AI security.
AI Research & Vulnerabilities:
Malicious MCP Servers Can Hijack LLM Hosts
Researchers discovered that malicious Model Context Protocol servers can take control of LLM hosts, manipulate model behavior, and exfiltrate data while evading current detection tools. The study identified multiple attack types across configuration and prompt components, exposing a blind spot in how LLMs connect to external systems.
AI Detects Hardware Trojans in Computer Chips
Researchers at the University of Missouri developed an AI-based system called PEARL that uses large language models to identify hardware trojans with approximately 97% accuracy. The model not only flags malicious logic but also explains why it's suspicious, helping engineers secure chips faster.
Research on Watermarking AI-Generated Images
Researchers at Queen's University examined how diffusion models can embed invisible watermarks during image generation to trace origin and authenticity. The study compared fine-tuning and noise-based methods, highlighting watermarking's role in securing AI outputs and its ongoing challenges with removal and forgery attacks.
Cybersecurity News
Capita Fined £14M for 2023 Data Breach Failures
The UK Information Commissioner's Office fined Capita £14 million after a 2023 cyberattack exposed the personal data of 6.6 million people. Investigators found the company failed to contain the breach for 58 hours and lacked basic safeguards such as privilege separation, timely alert response, and regular penetration testing. The ICO said the incident could have been prevented with proper controls, underscoring that "no organization is too big to ignore its cybersecurity responsibilities."
Experian Netherlands Fined €2.7M for Unlawful Data Use
The Dutch Data Protection Authority fined Experian €2.7 million for violating GDPR by collecting and processing personal data from public and private sources without consent. The agency found that Experian used this data to generate credit scores that affected consumer rates and deposits without informing individuals. Experian has ceased operations in the Netherlands and will delete its database by year-end.
Prosper Data Breach Impacts Over 17 Million Accounts
Peer-to-peer lending platform Prosper disclosed a data breach where attackers accessed its databases and exfiltrated sensitive user information, including Social Security numbers and financial details. Have I Been Pwned reports that 17.6 million accounts were affected, with data such as names, emails, and income levels exposed. Prosper says customer funds remain secure and plans to offer credit monitoring once the investigation concludes.
Nation-State Breach of F5 Networks
A suspected China-linked threat actor maintained long-term access to F5's internal engineering systems, stealing BIG-IP source code and internal vulnerability data. The breach persisted for nearly a year before detection. US federal agencies were ordered to identify and patch all F5 BIG-IP and F5OS devices by October 22, 2025, citing "imminent risk of exploitation."
Clop Targets Oracle E-Business Suite, Leaks Envoy Air Data
Envoy Air, a regional carrier owned by American Airlines, confirmed a breach in its Oracle E-Business Suite system after the Clop extortion group listed American Airlines on its leak site. The airline says no sensitive or customer data was affected, though limited business information was exposed. The attack is part of Clop's wider campaign exploiting Oracle EBS zero-days.
Google Unveils AI-Powered Cybersecurity Features
For Cybersecurity Awareness Month, Google announced several AI-driven security initiatives aimed at countering AI-enabled threats and strengthening digital defense. Highlights include CodeMender, an autonomous AI agent that uses Gemini models to identify and patch code vulnerabilities with self-validation; a new AI Vulnerability Reward Program centralizing bounties for AI-related flaws; and the expansion of the Secure AI Framework (SAIF 2.0) to cover emerging agentic-AI risks with new guidance and an industry-shared risk map.
Startup/VC News
Resistant AI Raises $25M Series B
Prague-based Resistant AI secured $25 million in Series B funding to scale its AI-powered fraud and financial crime prevention platform. The company uses machine learning to detect forged documents, synthetic identities, and money laundering.
European startup HyperBunker raised €800K in seed funding to scale production of its air-gapped anti-ransomware device that offers immutable, software-free local data vaulting for rapid recovery without paying ransoms.
That's all for now… Stay informed and protected.