AI Operator Agents Helping Hackers Generate Malicious Code

Symantec’s Threat Hunter Team has demonstrated how AI agents like OpenAI’s Operator can now perform end-to-end phishing attacks with minimal human intervention, marking a significant evolution in AI-enabled threats.

A year ago, Large Language Model (LLM) AIs were primarily passive tools that could assist attackers in creating phishing materials or writing code.

Now, with the introduction of agents such as OpenAI’s Operator (launched as a research preview on January 23), these AI systems can actively interact with web pages and perform complex tasks autonomously.

Symantec researchers successfully prompted Operator to identify a specific employee, determine their email address, create a malicious PowerShell script, and send a convincing phishing email, all with minimal guidance.

While their first attempt was blocked by the AI’s safety measures, they bypassed these restrictions by claiming authorization to send the email.

Increasing Sophistication and Future Risks

The demonstration revealed the operator’s concerning capabilities.

It successfully deduced an employee’s email address that wasn’t publicly available, drafted a PowerShell script after researching how to create one, and composed a convincing phishing email without requiring proof of authorization.

Although current agent technology remains relatively basic compared to skilled human attackers, the rapid advancement in this field suggests more sophisticated threats are on the horizon.

Researchers warn that future iterations could enable attackers to simply instruct an agent to “breach Acme Corp,” leaving the AI to determine and execute optimal attack strategies, including writing executables, establishing command-and-control infrastructure, and maintaining persistent network access.

This evolution significantly lowers the barrier to entry for potential attackers.

While agents like Operator may ultimately enhance productivity for legitimate users, they simultaneously open new avenues for exploitation that security professionals must prepare to counter.

The demonstration highlights the dual-use nature of AI agent technology and underscores the need for robust security measures as these tools become more powerful and accessible.

Are you from SOC/DFIR Teams? – Analyse Malware Incidents & get live Access with ANY.RUN -> Start Now for Free.