Follow us on Google News
A European telecommunications provider was reportedly targeted in a cyberattack lasting more than two weeks, with an artificial intelligence agent performing much of the intrusion. The operation, known as Operation Snowball, exploited security vulnerabilities and reached systems associated with 13 organisations, raising concerns about the growing use of autonomous AI tools in cybercrime.
Cybersecurity researchers discovered evidence suggesting that attackers used a customised AI system to identify vulnerable networks, execute attacks and move between connected systems with limited human intervention.
How Did the AI Agent Carry Out the Cyberattack?
Investigators found that the attackers had developed an autonomous framework using DeepSeek Flash, an artificial intelligence model capable of performing technical tasks based on instructions.
https://the420.in/fcrf-certified-chief-ai-officer-second-cohort-november-2026/
The human operator reportedly provided broad instructions in Chinese, allowing the AI agent to determine how to investigate systems, exploit weaknesses and continue its activities.
Researchers discovered approximately 70 records of interactions between the operator and the AI system in an exposed online directory. The records showed the agent testing different methods and modifying its approach when security controls prevented access.
Investigators also identified a jailbreak prompt containing approximately 78,000 characters, apparently designed to bypass the model’s safety restrictions and enable offensive cybersecurity activities.
The operation was not entirely autonomous. A human developed the framework and supplied instructions, while the AI system handled substantial portions of the technical work.
Researchers said the activity continued for more than two weeks, demonstrating how AI-assisted attacks could reduce the manual effort required for complex intrusions.
Which Security Vulnerability Allowed the Attack?
One significant entry point involved CVE-2026-20320, a high-severity vulnerability affecting Cisco BroadWorks.
The flaw involves XML External Entity processing in the Open Client Interface XML parser. It could allow an unauthenticated remote attacker to access sensitive files and configuration information by sending a specially crafted XML message.
Cisco disclosed the vulnerability on August 19, 2026, assigning it a severity score of 7.5. Security updates were made available, although the company indicated that no alternative workaround existed.
On November 10, the US Cybersecurity and Infrastructure Security Agency added the vulnerability to its Known Exploited Vulnerabilities catalogue, requiring federal civilian agencies to apply fixes by November 24.
Investigators said the attackers also exploited weaknesses in older enterprise applications to extend their access.
Were Millions of Healthcare Records Compromised?
The investigation raised particular concerns about a healthcare organisation hosted on the telecommunications provider’s infrastructure. Researchers reported that the attackers reached a TrakCare hospital information system connected to an Oracle database that reportedly retained a default password.
The database was believed to contain millions of patient records, potentially exposing sensitive healthcare information. Other information at risk reportedly included subscriber billing records, authentication credentials and private keys.
However, the findings did not establish how many records were actually downloaded or whether the entire database was compromised.
The researchers also noted that the operator’s instructions and the AI agent’s reports were written in Chinese. They suggested a possible Chinese connection but did not identify a specific hacking group or establish government involvement.
The incident highlights the risks posed by AI-assisted intrusions, particularly where known software vulnerabilities, weak passwords and inadequate access controls remain unresolved.
Security specialists have advised organisations to install available patches, strengthen authentication, restrict AI agent permissions and monitor unusual network activity to reduce exposure to similar attacks.
