A hacker with limited technical skills recently leveraged AI-powered frameworks to breach at least two Indian universities, making off with over 5,200 records from one of them. The attacker, using the alias BUGSTER, combined DeepSeek V4 with other tools to carry out data theft and reconnaissance efforts that would previously have required more experience.
How the Breach Unfolded
Investigators from Palo Alto Networks’ threat intelligence unit, Unit 42, discovered that the intruder had used AI to automate several stages of the attacks. The toolkit included Hermes Agent as an AI assistant framework, DeepSeek V4 for reasoning, and Cursor for writing code. Crucially, safety safeguards were disabled, allowing the AI to generate scripts and carry out research and reports without restriction.
In one university breach, the attacker mapped out infrastructure—WordPress site, student portal, admin dashboard—and was able to gain access by guessing a password. However, that particular mapping did not lead to the database theft. In a second case, using a weakly protected administrator account, the hacker accessed a MongoDB database and stole records containing details of students (diploma, undergraduate, postgraduate) and faculty. Details about the cloud and database setup of the target were also exposed.
Multiple Schemes and Tools Revealed
The intruder didn’t stop at university data-stealing. According to the investigation, they engaged in credential theft, tried cracking Wi-Fi passwords (requiring proximity to collect authentication data), and explored schemes like VoIP fraud and telecom carrier manipulation. Alongside, the hacker collected information on major cryptocurrency holders—but no follow-up attacks were found.
Besides obvious brute-force or weak-password attacks, other tools included credential-scanning kits that scanned for exposed cloud keys or secrets in public files and logs. AI-generated reports organized the research and automated script generation for tasks.
Lessons in Prevention
This case illustrates how attackers with modest hacking skills can significantly amplify their capabilities using AI agents—especially when core security practices are lax. Key takeaways include:
- Ensure strong, unique passwords for administrator accounts.
- Require multi-factor authentication for sensitive access.
- Limit database and cloud access strictly to authorized systems.
- Audit and remove unnecessary exposure of internet-facing dashboards or APIs.
- Protect secrets (API keys, credentials) by avoiding storage in plaintext or public files; rotate or revoke if exposed.
- If using AI agents internally, maintain safety controls, limited permissions, and regime for action reviews.
Other technical signals uncovered include IPs linked to infrastructure hosting several attacker-controlled services, and use of commercial proxy servers. No active telecom fraud was confirmed, just preparatory traces.
The broader picture is a rising threat: AI-augmented attacks are lowering the barrier for criminal campaigns. These incidents highlight weaknesses in basic security hygiene more than exotic vulnerabilities or zero-days. Organizations, especially educational institutions, are particularly exposed due to multiple portals, varied credentials, and often loose oversight.
