Entities

Entities collected from this article, grouped for quick reference and copying. Inclusion does not establish malicious activity. Article links appear separately at the bottom as references.

According to data Check Point published in its August 2026 cyber threat landscape report, ransomware has nearly doubled this year. The most interesting number in the report is not about threat actors. It is about a company’s own employees, and what they paste into chatbots like Claude, ChatGPT, Grok, and more.

The attack trend is ugly and clear. Organizations faced an average of 2,422 cyber attacks per week in August, up 22 percent year-over-year. Ransomware hit 1,042 attacks for the month, close to double the same month in 2025. Phishing crept up too, and roughly 1 in every 112 emails is now out to get you. The external threat is not slowing down. It is relentlessly compounding.

Check Point also tracks how enterprises use generative AI, and the finding is easy to misread. Of organizations regularly using GenAI, 86 percent were hit by high-risk prompt activity. It may sound like an attack statistic, but it is not.

The risk here is not threat actors using AI, and it is not flaws in the models. It is what employees actively type into the prompt. Sensitive customer records. Confidential financial data. Confidential legal documents. Infrastructure details. HR files. Check Point saw network and IT infrastructure data in 67 percent of affected organizations, financial data in 65 percent, legal and regulatory material in 64 percent, HR data in 59 percent, and PII in 57 percent. All of it was copied straight into third-party AI tools sitting outside the traditional security boundary of the company.

The counterintuitive part is the rate of risky prompts actually fell in August, to 1 in 43, the lowest in months. So people are getting slightly more careful per prompt. Yet the exposure still spread to 86 percent of AI-using organizations, because usage exploded.

The average user generated 106 prompts in August, up from 78 in June, across an average of seven different AI tools per company. The per-prompt risk dropped, and the total risk went up anyway. People are being slightly more careful per prompt, but they are prompting far more often, which adds up to more exposure.

Everyone is watching the ransomware number climb, and they should, because it is treacherously high. But one of the fastest-growing ways sensitive corporate data leaves the building right now is not a threat actor breaking in and exfiltrating documents. It is an employee trying to be helpful. Someone pastes a contract into a chatbot to summarize it. Someone else drops a customer list in to clean it up. The data now sits in a third party’s system, outside every control the company spent money to build.

It does not feel like a breach, which is exactly why it is dangerous. A ransomware attack announces itself. Careless AI use leaves no alarm, no ransom note, and no incident to respond to. Just a slow, invisible drip of the most sensitive data a company holds into tools nobody is monitoring, one helpful prompt at a time. The healthcare sector is worst, with 1 in every 25 prompts carrying sensitive data. Think about what “sensitive” means in healthcare.

The source deserves an honest note. This is Check Point data, and Check Point sells prevention. The framing naturally points toward buying more of its security tooling. The numbers are credible and match what others are seeing. Just exercise critical thinking, and read any vendor threat report knowing it doubles as a sales pitch.

The takeaway is not to ban the tools. The ship sailed, and in any event the productivity is real. Most companies deployed generative AI to their whole workforce before they built a single rule about what could go into it.

The attack surface everyone watches is the network. The one quietly leaking the most right now is the text box.

Companies spent decades building walls around their data. Then they put a blinking cursor inside those walls and invited every employee to carry pieces of the company through it.

The new exfiltration channel has a send button.