A new cybersecurity report has highlighted how generative artificial intelligence is increasingly being incorporated into offensive cyber operations. Researchers allege that a Russian-speaking threat actor misused Google's open-source Google Gemini CLI as a "hacking agent" to streamline botnet administration, automate infrastructure migration, and assist with various operational tasks through natural-language prompts.

The findings illustrate how AI-powered developer tools, originally designed to improve productivity, can also be abused to accelerate malicious cyber activities.

More Than 200 AI Sessions Recorded

According to the report, the threat actor—identified as "bandcampro"—interacted with Gemini CLI in over:

• 200 AI sessions
• Conducted during May and June 2026

Researchers recovered operational logs indicating that:

• AI provided technical troubleshooting at least 59 times
• Generated code
• Assisted operational planning
• Suggested workflow improvements
• Helped resolve infrastructure issues

AI Used to Operate Small Botnet

The report alleges the attacker used Gemini CLI to manage a small botnet controlling:

• 8 compromised computers
• Located at a dental clinic

Researchers claim the attacker also attempted to access:

• OpenDental databases
• Internal systems
• Administrative infrastructure

To bypass safety restrictions, the attacker reportedly instructed Gemini CLI to behave as an "authorized penetration tester," allowing responses without triggering certain standard safety prompts.

AI Generated Complete C2 Migration Plan

One of the report's most significant findings involves the migration of the botnet's command-and-control (C2) infrastructure.

Researchers state that after receiving a simple instruction—

"Study the C2 migration"

—the AI reportedly generated:

• Server deployment code
• VPS configuration
• Cloudflare Tunnel setup
• Initial debugging procedures
• Infrastructure migration plan

According to researchers, the migration process was completed in approximately:

• 6 minutes

AI Helped Restore Botnet Connectivity

Following the migration, several infected systems initially failed to reconnect.

The report states that Gemini CLI identified:

• Traffic conflicts
• Old server interference
• Configuration inconsistencies

Once the previous infrastructure was disabled, researchers claim:

• All compromised systems successfully reconnected
• Botnet operations resumed normally

Natural-Language Management of Compromised Systems

Investigators allege the attacker administered the botnet almost entirely through conversational prompts.

Reported AI-assisted tasks included:

• Identifying active infected systems
• Listing files on compromised machines
• Creating new infection links
• Troubleshooting server issues
• Managing operational workflows

The report notes that the operational framework consisted of only:

• 3 plain-text files
• Total size approximately 5 KB

These reportedly included:

• Gemini jailbreak prompt
• Command-and-control playbook
• Migration instructions

Malware Described as Technically Simple

Researchers observed that the malware itself was not highly sophisticated.

The reported toolkit included:

• In-memory Python HTTP server
• PowerShell agents
• Windows Scheduled Tasks
• WMI Event persistence
• Registry-based persistence

Unlike many advanced malware families, researchers found little evidence of:

• Heavy obfuscation
• Complex packing
• Advanced anti-analysis techniques

AI Also Assisted Credential Attacks

According to the report, the attacker additionally used Gemini CLI to:

• Generate WordPress password combinations
• Analyse 1Password data dumps
• Search for exploitable credentials

However, researchers state that Gemini CLI refused one request to generate a self-propagating "agent bomb," after which the attacker reportedly shifted to other activities.

AI Increasingly Becoming a Cybercrime Force Multiplier

Cybersecurity experts believe the incident highlights an important evolution in cyber threats.

Rather than writing malware independently, attackers can increasingly leverage AI to:

• Automate repetitive tasks
• Accelerate troubleshooting
• Produce infrastructure code
• Simplify command execution
• Reduce technical barriers
• Improve operational efficiency

While AI does not replace attacker expertise, it can significantly increase productivity.

Experts Call for Stronger AI Safeguards

Cybercrime expert and former IPS officer Prof. Triveni Singh observed that generative AI is rapidly evolving beyond productivity applications.

According to him, if adequate safeguards are not implemented, AI-powered automation tools may substantially increase the speed and effectiveness of cybercriminal operations.

He emphasized the need for:

• Continuous AI monitoring
• Strong misuse detection
• Behavioral safeguards
• Security-by-design principles
• Responsible deployment of AI-powered developer tools

Growing Challenge for AI Security

Researchers say incidents like this demonstrate a growing cybersecurity challenge.

As AI coding assistants become deeply integrated into software development and IT operations, organizations will need stronger governance mechanisms to prevent malicious misuse while preserving legitimate productivity benefits.

Future security frameworks are expected to increasingly focus on:

• AI activity monitoring
• Prompt abuse detection
• Behavioral analytics
• Access controls
• Infrastructure auditing
• Human oversight

Conclusion

The research highlights how generative AI tools can be repurposed beyond their intended use when placed in the hands of threat actors.

Although the report describes a relatively small botnet, it demonstrates how AI can assist with infrastructure deployment, troubleshooting, automation and operational planning—potentially reducing the technical effort required to conduct cyber operations.

As AI capabilities continue to expand, balancing innovation with robust security controls will become increasingly important for developers, enterprises and AI platform providers alike.

Shunyatax Global Insight

Shunyatax Global says that generative AI is rapidly becoming a dual-use technology. While it dramatically improves developer productivity, it can also accelerate cybercriminal operations when safeguards are bypassed. Organizations deploying AI coding assistants should implement strict access controls, monitor prompt activity, restrict sensitive infrastructure access, and continuously audit AI-assisted workflows. AI governance should now be considered a core component of enterprise cybersecurity strategy rather than an optional compliance measure.