AI-Assisted Hacker Stole 5,200+ Records From Indian University, Unit 42 Reports

Artificial intelligence is changing the cybersecurity landscape, and a recent investigation shows how attackers can combine AI assistants with conventional tools to support cybercrime. Palo Alto Networks’ Unit 42 reported that a threat actor using the online alias BUGSTER targeted organizations in India, including universities, and stole a database containing more than 5,200 student and staff records in one confirmed intrusion.

According to the investigation, the actor used Hermes Agent, DeepSeek V4 and Cursor to assist with reconnaissance, scripting and research. The findings highlight how AI can lower some technical barriers for inexperienced criminals, although the attacks still involved a human operator directing the tools.

How the Attacks Targeted Indian Universities

Unit 42 investigated activity involving two universities. In one case, the actor used AI-assisted research to map a university’s online education platform and related services, including a WordPress website, student portal and administration dashboard.

The actor then attempted account passwords and identified a working login. However, Unit 42 did not establish that this activity caused the separate database theft.

In another incident, the actor accessed a university database using an administrator account protected by a weak password. The stolen MongoDB database contained more than 5,200 records, including personally identifiable information belonging to students and staff.

The incidents demonstrate why organizations must secure internet-facing applications, protect administrator accounts and restrict access to sensitive databases.

Figure : Excerpt from a WordPress REST API enumeration report associated with the activity investigated by Unit 42. Source: Unit 42.

How AI Tools Supported the Threat Actor

The investigation found that BUGSTER configured Hermes Agent with DeepSeek V4 as a reasoning model and used Cursor as an additional coding tool. Unit 42 reported that the actor relied on AI-generated scripts and reconnaissance reports to organize technical work.

Researchers also identified activity involving credential theft, automated Wi-Fi password cracking, cryptocurrency research and preparations for potential telephone-service fraud.

These findings require careful interpretation. Investigators found planning material related to VoIP fraud, but did not find call records or evidence establishing that the planned telephone operation became active. Similarly, researching cryptocurrency holders does not, by itself, prove that those individuals were subsequently attacked.

The wider lesson is that AI assistants can help a single operator organize multiple tasks, but AI-assisted activity should not automatically be treated as fully autonomous cybercrime.

Evidence of Broader Cybercrime Planning

Unit 42 also found a credential-scanning toolkit designed to search for exposed secrets, including cloud keys and private keys in files or logs. Exposed credentials can create opportunities for unauthorized access when organizations fail to protect them properly.

The researchers described a separate AI profile used to investigate telephone carriers and seek information about possible VoIP fraud.

Figure : Screenshot illustrating carrier-related research documented in the Unit 42 investigation. Source: Unit 42.

The report also discusses Wi-Fi password-cracking activity that required physical proximity to targeted networks. This is an important distinction: the capability described was not simply a remote attack against every nearby wireless network.

Security Recommendations for Universities

Universities and other organizations should focus on preventing the weaknesses highlighted by this investigation:

• Strengthen authentication: Require multifactor authentication for administrator accounts and eliminate weak or reused passwords.
• Restrict database access: Allow database connections only from authorized systems and users.
• Review public-facing services: Check websites, portals and APIs for unnecessary exposure of account information.
• Protect credentials: Keep API keys, cloud credentials and private keys out of public repositories, files and logs.
• Monitor suspicious activity: Investigate unusual login attempts, unexpected administrator access and large-scale database queries.
• Control AI agents: Apply least-privilege permissions, retain appropriate safety controls and review actions performed by AI tools.

These measures address the underlying security risks regardless of which AI model an attacker uses.

Conclusion

The Unit 42 investigation shows how AI-assisted research and scripting can support a range of cybercrime activities, including confirmed university intrusions and the theft of sensitive records.

For universities, the priority should be securing administrator accounts, limiting database exposure and monitoring access to personal information. For organizations adopting AI agents, careful permission management and oversight are equally important.

Source: Palo Alto Networks Unit 42 — “Next Exploit Opportunities”: How a Threat Actor Used AI to Help Plan and Execute Multiple Types of Cybercrime Activities.