Cybersecurity experts are raising red flags as hackers rapidly adopt AI tools to launch smarter, faster attacks.
According to a new report by Check Point Software Technologies, defenders need to step up and use AI themselves to fight back.
Presented at the 2025 RSAC Conference in San Francisco, the report highlights an alarming trend: one in 13 generative AI prompts involves sensitive content, and one in every 80 is at high risk for serious data exposure.
Top threats include unauthorized AI use, data leaks, and weak points in AI platforms.
These risks put companies in danger, especially those that haven’t taken AI seriously yet.
Key Takeaways:
Check Point’s findings offer a wake-up call for businesses. While AI can streamline tasks, it also opens doors for risks if used without proper controls.
If employees use AI tools without permission, companies may face security gaps, legal problems, and mishandled data.
The report stresses that unapproved AI apps can trigger major issues — from system vulnerabilities to breaches. Businesses are urged to evaluate AI tools before rolling them out.
Some platforms may store or share data insecurely, increasing the chances of a cyberattack.
AI in the Hands of Hackers:
Researchers discovered that hackers are already using ChatGPT and OpenAI’s APIs, while tools like Microsoft Copilot, Google Gemini, and Anthropic Claude are also gaining traction in dark web circles.
Meanwhile, open-source models such as Alibaba’s DeepSeek and Qwen are entering the scene with few restrictions, making them ripe for misuse.
Even worse, criminal groups have begun crafting AI tools specifically for attacks — like WormGPT and an AI-powered DDoS system from the FunkSec ransomware group.
A Changing Battlefield:
Check Point urges defenders to adopt AI-powered tools — including their own anomaly detection systems — to stay ahead of these evolving threats.
The report is clear: AI in cybercrime isn’t just a theory anymore. It’s real, it’s active, and it’s advancing faster than many companies’ current security systems can handle.
