Summary

• Check Point recorded 2,422 cyber attacks per organisation per week globally, up 22% year on year.
• Europe recorded the fastest regional growth at 28%, while enterprise users generated an average of 106 GenAI prompts during August.
• Check Point counted 1,042 ransomware victims from criminal leak sites, almost twice August 2025, although the dataset is not a complete census of incidents.

Cyber attacks against European organisations increased faster than in any other major region during August, while growing use of generative AI created more opportunities for sensitive company information to move into tools outside traditional security controls, according to new data from Check Point Software Technologies.

The security company recorded an average of 2,422 attacks per organisation each week globally, 4% above July and 22% higher than a year earlier. Europe recorded the fastest annual growth at 28%, even though other regions still experienced greater attack volumes in absolute terms.

Visible ransomware activity also increased sharply. Check Point identified 1,042 victims disclosed through ransomware groups’ public leak sites during August, almost twice the number seen in August 2025 and 8% above July.

That figure needs qualification because it is based on criminal disclosure sites rather than every ransomware incident. Attacks settled before publication, victims never named by a group, and campaigns without a leak-site model can all be absent, making the dataset a measure of observable extortion activity rather than a complete global tally.

More AI use creates more opportunities for leakage

Check Point’s GenAI data shows a different type of security pressure developing alongside conventional attacks. The average enterprise user generated 106 prompts during August, compared with 95 in July and roughly 78 in June, while organisations in the dataset used seven AI tools on average.

The proportion of prompts classified as presenting a high risk of sensitive-data exposure actually fell, reaching one in every 43 prompts. Even so, 86% of organisations using GenAI regularly experienced some high-risk prompt activity.

The combination suggests that organisations may be improving controls or employee behaviour while the overall amount of AI interaction continues to grow. A lower percentage of risky prompts can still produce substantial exposure when employees send more requests through more tools each month.

Healthcare and medical organisations had the highest high-risk prompt rate at 4%, followed by software at 3.6% and business services at 3.5%. Those sectors routinely handle information that can include regulated personal data, intellectual property, source code, financial information, or commercially sensitive material.

Unlike a conventional data-loss route, GenAI use can cross several security boundaries at once. Employees may switch between approved enterprise systems, browser-based services, embedded assistants, and specialist tools, while organisations do not always know which model is receiving a prompt, what the service retains, or whether the information falls within an approved processing arrangement.

As AI agents acquire permission to retrieve data and act on users’ behalf, the control problem becomes less dependent on whether an employee manually copies information. Systems capable of reading mailboxes, shared drives, customer records, or code repositories can move data between services automatically unless permissions and policy enforcement travel with the workflow.

Older attack methods remain effective

While AI introduces a newer route for data exposure, Check Point’s figures show phishing remaining stubbornly conventional. One in every 112 emails was classified as phishing during August, compared with one in 128 in July, while links appeared in 72% of phishing messages and attachments in 14%.

The dominance of links reflects the economics of social engineering. Attackers can redirect users towards credential-harvesting pages or malware infrastructure without having to attach a malicious file, while links can be adapted quickly around invoices, account warnings, tenders, travel, or other ordinary business activity.

Education experienced the highest overall attack volume in Check Point’s dataset, with 5,354 weekly attacks per organisation, 28% above a year earlier. Government followed with 3,067, while hospitality, travel, and recreation reached 3,056, an annual rise of 56%.

Ransomware disclosures were concentrated in business services, which accounted for 36% of visible victims, followed by industrial manufacturing at 13% and consumer goods and services at 12%. Service providers can be attractive because disrupting one organisation may affect multiple customers or business processes farther along a supply chain.

Qilin was the most active ransomware operation in the dataset at 15% of published victims, followed by The Gentlemen at 10%, while Orova entered the top three at 4%. Such rankings change quickly as affiliates move between operations, groups rebrand, or law-enforcement action disrupts infrastructure.

Security controls are converging around data

The August figures describe an enterprise environment where new AI risks are accumulating on top of established attack routes rather than replacing them. Phishing still captures credentials, ransomware still exploits operational disruption, and vulnerabilities remain valuable to attackers, while employees and software agents are simultaneously gaining more ways to move information through AI services.

That makes separate security policies for email, SaaS, cloud, endpoints, and AI increasingly difficult to operate independently. The same sensitive customer record can move through several of those environments during one workflow, while an organisation may discover that each security product sees only part of the path.

The practical response is therefore likely to centre increasingly on identity, classification, access, and control of the data itself rather than treating GenAI as a wholly separate threat category. Organisations still need to decide which AI tools are acceptable, but they also need controls capable of following sensitive information when it moves between ordinary software and AI-assisted processes.

Check Point’s figures cannot establish how much actual loss results from risky prompts, nor do they prove that rising GenAI use caused the wider increase in cyber attacks. They do, however, show the two trends occurring inside the same operating environment: attack volume is climbing while employees are sending more information through a growing collection of AI tools. Security teams increasingly have to manage both without assuming one will wait for the other.