This Week in AI Security - 18th D...

This Week in AI Security - 18th December 2025

Modern Cyber with Jeremy Snyder by Jeremy Snyder
S1 · E83
Dec 18, 2025
10:47

Episode notes

In the final episode of 2025, Jeremy examines the evolution of SEO poisoning into "AI poisoning," a major privacy breach involving a popular browser extension, and shares a data-driven "sneak peek" at the state of AI security over the past year.

Key Stories & Developments:

  • AI Poisoning of Search Results: Researchers identified an attack where threat actors plant false information online to trick AI-powered search engine crawlers. This results in search engines providing AI summaries that list scam phone numbers for legitimate services like airline call centers, effectively creating a modern, AI-driven version of SEO poisoning.
  • The "Pay-to-Crawl" Proposal: Jeremy discusses a new proposal from Creative Commons that suggests moving away from outright blocking AI crawlers. Instead, website owners could set a price for crawling and training, allowing organizations to monetize the use of their data by LLM providers.
  • Urban VPN Privacy Breach: A popular Chrome and Edge extension, Urban VPN Proxy, was caught intercepting and reading the AI chat messages of its 7.3 million users. This incident highlights the risk of third-party browser extensions reading sensitive data that users assume is private.

2025 in Review Snapshot: Using data from the Firetail AI Incident Tracker, Jeremy reveals two major trends from 2025:

  • The Surge in Incidents: AI security incidents saw a massive jump from 2024 to 2025, marking this as the year AI-related security became a global, pervasive problem.
  • Disclosure vs. Injection: While the OWASp Top 10 lists prompt injection as the #1 risk, the tracker data shows that sensitive information disclosure (largely due to organizational error) actually outstrips prompt injection by about a third.

Episode Links

Keywords

AI
AI Security
AI Governance
This Week in AI Security
AI News