We developed Scarecrow, a tool that hides refusal instructions in documents so AI agents ingesting them are prompted to stop processing. Our findings show that this same technique can help defenders disrupt AI-driven access to sensitive documents.
Generative AI now writes clinical notes, drafts patient replies, and decides claims, often faster than healthcare’s oversight can keep pace. TrendAI™ Research maps the resulting “custody gap” and the risk it creates, from the exam room to the back office, where AI-generated artifacts can enter patient records.
In this report, we present the security and governance issues enterprises should consider when adopting and using AI agents. We lay out the risk structure unique to AI agents and present a security framework and a method for technical and organizational controls to support AI adoption.
Our TrendAI™ Research team developed AIxploit, a framework that systematically maps an agent’s AI-native attack surface through attack testing.
We developed Scarecrow, a tool that hides refusal instructions in documents so AI agents ingesting them are prompted to stop processing. Our findings show that this same technique can help defenders disrupt AI-driven access to sensitive documents.
Generative AI now writes clinical notes, drafts patient replies, and decides claims, often faster than healthcare’s oversight can keep pace. TrendAI™ Research maps the resulting “custody gap” and the risk it creates, from the exam room to the back office, where AI-generated artifacts can enter patient records.
In this report, we present the security and governance issues enterprises should consider when adopting and using AI agents. We lay out the risk structure unique to AI agents and present a security framework and a method for technical and organizational controls to support AI adoption.
Our TrendAI™ Research team developed AIxploit, a framework that systematically maps an agent’s AI-native attack surface through attack testing.
How AI is accelerating traditional security failures, and why familiar threats are now harder to detect and contain.
We examine how OpenClaw’s exposed deployments, plaintext credential storage, prompt injection risks, unguarded MCP integrations, and malicious third-party skills amplify the security risks of agentic AI at enterprise scale.
The critical systems we can’t afford to lose are the same ones attackers are racing to target. With a new generation of AI-augmented cyber threats, what used to be weeks of warning can collapse into mere hours. What is the current state of the threat landscape for essential industries, and what can organizations do to fortify their defenses? Uncover this and more in our report.
China-aligned operational relay box (ORB) infrastructure is harvesting open sensor data at scale, and most defenders can't see it. Our report dives into how these ORBs can use open telemetry for data collection and other purposes.
TrendAI™ Research went inside the forums, marketplaces, and Telegram channels where access to factories, utilities, and energy firms is bought, sold, and weaponized. Combing through two years’ worth of data revealed an underground where financially motivated brokers and ransomware crews now operate alongside state-aligned hacktivists, sharing the same entry vectors, the same pirated training, and in some cases, the same personnel.
What happens to an unpatched or vulnerable application or organization’s IT infrastructure? Here's how virtual patching helps enterprises address vulnerability and patch management woes.
30. September 2026
6 min