AI Operator Agents Assisting Hackers To Write Malicious Code

AI-powered agents are evolving rapidly, offering enhanced capabilities to automate routine tasks, but researchers have discovered these same tools can be weaponized by malicious actors. OpenAI’s Operator, launched as a research preview on January 23, 2025, represents a new generation of AI tools that can interact with web pages and perform complex tasks with minimal […] The post AI Operator Agents Assisting Hackers To Write Malicious Code appeared first on Cyber Security News.

Mar 17, 2025 - 14:45

AI Operator Agents Assisting Hackers To Write Malicious Code

AI-powered agents are evolving rapidly, offering enhanced capabilities to automate routine tasks, but researchers have discovered these same tools can be weaponized by malicious actors.

OpenAI’s Operator, launched as a research preview on January 23, 2025, represents a new generation of AI tools that can interact with web pages and perform complex tasks with minimal human intervention.

While designed for legitimate purposes, these agents could potentially be leveraged by attackers to create infrastructure and mount sophisticated attacks.

In a concerning demonstration, AI agents successfully completed multiple steps of a simulated attack with little human guidance.

The test showed how easily these systems could be manipulated to perform reconnaissance, create malicious code, and deliver it using social engineering techniques.

Symantec Security researchers noted that with simple prompt modifications, they could bypass AI safety guardrails.

In their experiment, researchers merely needed to claim authorization from a target before Operator willingly proceeded with potentially harmful actions, including identifying a specific employee, deducing their email address, and crafting a convincing phishing email.

The demonstration revealed how Operator independently researched PowerShell commands before writing a script designed to gather sensitive system information.

PowerShell script created by Operator (Source -Symantec)

The resulting code included commands to collect operating system details, computer specifications, network configuration, and disk information—all valuable intelligence for an attacker establishing a foothold in a network.

Perhaps most concerning was Operator’s ability to craft a convincing phishing email, where it impersonated an IT support professional named “Eric Hogan” and created a plausible pretext for the target to execute the script.

Email that Operator sent to target, Symantec’s Dick O’Brien (Source -Symantec)

The message urged execution to “ensure system integrity and performance” as part of “ongoing efforts”—language typical of legitimate IT communications.

Technical Implications

The PowerShell script created by Operator showcases how AI can now write functional malicious code without human expertise.

The script uses standard Windows Management Instrumentation (WMI) commands to extract system information and save it to a text file in the user’s profile.

While this particular example gathered only system information, the same approach could be used to create more damaging payloads.

Security experts warn that as these AI agents become more sophisticated, attackers could potentially instruct them to “breach Company X” and have the AI determine and execute optimal attack strategies automatically, dramatically lowering the technical barrier for conducting cyber attacks.

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

The post AI Operator Agents Assisting Hackers To Write Malicious Code appeared first on Cyber Security News.