AI Threats
Every TrendGuru field note filed under AI Threats — emerging threats, defensive patterns, and what is actually working in production.
4 articles
Agentic Browsers: When Your AI Assistant Has a Cursor
Perplexity Comet, Arc's Dia, OpenAI's Operator, Anthropic's Computer Use — 2026 is the year the assistant stopped answering and started clicking. The threat model changed with it.
Prompt Injection in 2026: The New Perimeter Attack
Indirect prompt injection has quietly become the most exploited weakness in enterprise LLM deployments. Here's what changed and how to defend.
MCP Server Sprawl: The New Attack Surface Nobody Inventoried
Model Context Protocol went from Anthropic curiosity to industry default in eighteen months. Now every developer laptop runs a dozen unaudited MCP servers with access to filesystems, APIs and databases.
Model Supply Chain Attacks: The npm Moment for AI
Poisoned model weights on public hubs are the new typosquatted packages. What supply chain security taught us translates almost directly.