Analysis · curated 12 Sep 2026

From Jailbreaks to Agentic Attacks: The Evolution of AI Abuse

Coverage timeline

12 Sep 2026bitsight.com

Single-source analysis — first reported, latest, and curated coincide.

Why it matters

Bitsight's research frames a shift defenders must track: AI abuse is maturing from prompt manipulation into repeatable agentic execution where MCP tool access turns unsafe model behavior into real-world security risk.

Bitsight Threat Intelligence published research titled "From Jailbreaks to Agentic Attacks: The Evolution of AI Abuse" tracing how threat actors are moving beyond copy-paste jailbreak prompts toward prompt injection, obfuscation, multi-model workflows, AI coding agents, and MCP-connected tool abuse. The report argues that in agentic environments what a model can do (access files, run commands, call enterprise tools) matters more than what it can say, expanding the impact of indirect prompt injection.