Attackers are using AI chatbots to write custom malware, and the throwaway nature of the output is becoming a real problem for defenders.
Security firm Huntress analyzed a PowerShell script called Untitled1.ps1 — a name that practically announces its origins — and concluded it was almost certainly generated by a generative AI tool. The script was a custom Active Directory enumeration tool: aggressive, loud, and messy in the way AI-produced code tends to be, complete with leftover comments the model forgot to clean up. The attackers paired it with s5cmd, a legitimate Amazon S3 command-line tool frequently repurposed for bulk data exfiltration, and SharpShares.exe, a known share enumeration utility. They were caught and evicted before finishing the job.
The real problem is not this particular script — it's that every AI-generated payload is, by definition, unique. Traditional antivirus and endpoint detection tools lean on file hashes and static string signatures. Untitled1.ps1 has never existed in that exact form before and will never exist again, which means signature-based defenses have nothing to match against. Huntress argues defenders need to shift toward behavioral analytics: the underlying mechanics of Active Directory enumeration do not change just because an LLM rewrote the syntax.
This is the logical endpoint of a trend that has been building since AI coding assistants went mainstream — the same tools that help developers ship faster are now shortcutting the skill floor for low-level cybercriminals. The attacker here was noisy enough to get caught, but the tooling will only get quieter.