Threat Actors Weaponized AI Tools to Gain Full Domain Access within 30 Minutes

Threat actors transformed popular artificial intelligence tools into weapons in 2025 so they could launch quick, accurate network intrusions This article explores 2025 speed intrusion. . According to CrowdStrike's 2026 Global Threat Report, attacks by AI-enabled adversaries increased by 89% year over year as hackers reduced the time between initial entry and full domain access to less than 30 minutes using automation and machine-generated scripts.

The most distinctive aspect of the threat landscape in 2025 was the speed of intrusion. In comparison to 2024, the average eCrime breakout time, or the time between obtaining initial access and moving laterally to other systems, decreased to 29 minutes, a 65% speed increase. It only took 27 seconds for the fastest breakout ever recorded.

In one case that has been documented, data exfiltration started within four minutes of initial access, giving organizations very little time to take action. Analysts at CrowdStrike pointed out that the techniques used to achieve this acceleration were closely related to misuse of AI. In addition to creating unique malware, adversaries were infecting trustworthy AI tools operating within victim environments with malicious prompts.

By inserting malicious JavaScript into Node Package Manager (npm) packages in August 2025, attackers were able to take control of victims' local AI tools, including Claude and Gemini, and steal cryptocurrency assets and authentication credentials. OverWatch and CrowdStrike Services addressed over ninety affected clients. One noteworthy instance concerned CHATTY SPIDER, an eCrime adversary that used voice phishing to target a U.S.-based law firm.

Through Microsoft Quick Assist, the team persuaded an employee to allow remote access. Within four minutes, CHATTY SPIDER attempted to use WinSCP to transfer stolen files to infrastructure under the control of the attacker. Within four minutes, CHATTY SPIDER begins to steal data (Source: Crowdstrike).

The attacker switched to Google Drive after the firewall blocked it. Before any data left the network, CrowdStrike OverWatch halted the exfiltration. Threat actors like FAMOUS CHOLLIMA constructed multi-phase attack pipelines with AI assistance in addition to individual operations. They created fictitious personas, managed numerous accounts, and carried out technical job tasks while using tools like ChatGPT, Gemini, GitHub Copilot, and VSCodium.

As a result of AI reducing the effort needed to conduct extensive deceptive operations, their 2025 activity doubled over 2024.

Threat Actors' Use of AI as a Weapon Throughout the Kill Chain Gemini-generated scripts were used by PUNK SPIDER, the most active ransomware adversary in 2025 with 198 recorded intrusions, to retrieve credentials from Veeam Backup & Replication databases. DeepSeek-generated scripts were probably used to stop services and destroy forensic evidence. 2024 vs. 2025 AI threats throughout the kill chain (Source: Crowdstrike) The LAMEHUG malware, which was used by Russia-affiliated actor FANCY BEAR, used hardcoded prompts to query the Hugging Face LLM Qwen2.5-Coder-32B-Instruct in order to conduct reconnaissance and gather documents prior to exfiltration.

This avoided static security tools by substituting AI-generated outputs for strict code logic. Remarkably, 82% of all 2025 detections were malware-free, indicating that the majority of attacks used approved channels rather than conventional malicious software.

To identify fast-moving intrusions before they spread, organizations should keep an eye on the use of AI tools on endpoints, patch AI platforms quickly, audit npm dependencies, and maintain cross-domain visibility across identity, cloud, and SaaS environments. Set ZeroOwl as a Preferred Source in Google and use X, LinkedIn, and LinkedIn to receive more real-time updates.