Ledger / Mar 2026
Go dropper asks GPT-4 whether to deploy Sliver
Confirmed
Unit 42's code analysis shows the dropper sends host data 'to OpenAI's GPT-4 model using an HTTP API' and runs Sliver only on the model's verdict. Malware that calls a model as it runs is direct evidence of AI use, so the grade is Confirmed. Unit 42 reports 'No campaign or victims identified'.
- Model
- ChatGPT / GPTOpenAI · closed
- Agent harness
- Nonechat or tool use only
- Attributed country
- Not attributedthe source names no country
- Sectors hit
- unknownWindows hosts of unknown victims
Unit 42 analyzed a Go dropper that sends host telemetry to GPT-4 and runs its Sliver payload only if the model says to execute, plus a .NET stealer whose GPT-3.5 calls were mostly theatre.
“Unit 42 analysed a Golang dropper that sends host telemetry to GPT-4 posing as an OPSEC AI and only runs its Sliver payload if the model returns execute true, plus a .NET stealer whose GPT-3.5 calls were mostly theatre.”AI Threat Watch
Sources
- Go dropper asks GPT-4 whether to deploy SliverAI Threat Watch · Mar 2026
- Analyzing the current state of AI use in malware: .NET infostealers integrating GPT-3.5-TurboPalo Alto Networks Unit 42 · Mar 2026
- Date
- Mar 2026
- Actor
- Unknown operators (unknown), as attributed by AI Threat Watch
- Targets
- Windows hosts of unknown victims
- What the AI did
- LLM inside malware, Evasion
- Model as named
- GPT-4
- Model basis
- A researcher names it
- Autonomy
- Assisted a human
- Outcome
- The Sliver payload ran only when the model approved.