Skip to content
by InfinyAI ES

IA en un minutoNewsSecurity

SecurityUpdated:

Check Point launches a feature to protect AI agents from booby-trapped documents

IA en un minuto newsroom · Editor: Jon Elgezabal

In 30 seconds

PromptGuardX, Check Point's new feature, scans every PDF for buried commands before an AI agent or copilot gets to read its contents. The tool exposes concealed layers, invisible text and character tricks, and flags the passage that set it off along with a confidence level. It sits inside Threat Emulation, and Check Point admits that more defenses are still required.

Check Point Software, a cybersecurity company, on Tuesday, October 6, presented PromptGuardX, a feature that searches PDF files for hidden instructions before their content is passed to an AI model, a copilot or an agent.

Its target is indirect prompt injection (a prompt being the instructions an AI is given): an attacker tucks commands into a PDF, an email or a web page and waits until an AI reads it. A person looking at the document sees nothing unusual, but the AI works with the text pulled out of the file's internal structure, which can hold invisible text, hidden layers or tampered Unicode characters. Check Point notes that a file can be entirely free of malware and still carry an order written for the agent.

The company's demonstration uses an ordinary-looking invoice with a concealed instruction inside. Asked to summarize it, the AI is pushed by that instruction to open the computer's calculator, a harmless stand-in for running a command. Since the invoice holds no executable and takes advantage of no flaw in the PDF format, nothing happens unless the AI reads the instruction, obeys it and can reach the appropriate tool. Inside a company, Check Point explains, a comparable instruction might try to pull out supplier data, send information outside, alter payment details or call a tool that has nothing to do with what the user asked for.

PromptGuardX examines every PDF in four steps. First it extracts the text the AI would receive, hidden text included, and normalizes the characters to defeat disguise tricks. Next it pinpoints suspicious areas, such as passages posing as system instructions or asking for earlier instructions to be ignored. Then a classifier trained to tell text that talks about an instruction apart from text that tries to give one weighs the context. Finally it delivers a verdict, with a confidence score and the fragments of text that prompted it.

The feature is part of Threat Emulation, the Check Point system that inspects files for malware, malicious links and suspicious behavior, and is available in supported deployments of Check Point Firewall, of Check Point Email Security and of gateways with Threat Emulation switched on. No price appears in the announcement. Check Point itself concedes that no control wipes out this type of attack single-handedly, and that other safeguards remain necessary: inspecting prompts, monitoring agents while they run, granting them only the permissions they strictly need and enforcing controls on each action.

Its AI Security Report 2026 says detections of long malicious instructions grew roughly fivefold from March to May 2026, to nearly 1% of the prompts observed in May.

Correction (12 October 2026): the products where it is available are now named as Check Point names them in its announcement.

Why it matters · analysis and opinion

Until now, checking a file meant asking whether it carried malware. With agents that read invoices and other documents on their own, the question becomes whether the text is trying to give them orders, and a cybersecurity vendor building that into its file inspection shows the risk has already left the lab. That said, for now it analyzes PDFs and works inside Check Point's own products, so it does not replace the basics. Anyone who already lets an AI assistant read outside documents and then act can start today with three things: give it minimal permissions, require a person to confirm payments or data transfers, and keep document reading separate from the tools that carry out actions. The filter sits in front of the agent, but the damage depends on what the agent is allowed to do.

Official source: Check Point · Written with the help of AI: how we make the news

AI in your inbox, every day or every Friday

The stories that matter, each one in a minute. With the source for every one.

Choose one or both:

Sign up and you are in: the daily arrives every night and the weekly on Friday mornings. You can unsubscribe from any email.