Threat video1:00
Kerkoporta Coding Agent Backdoor: Detection & IOCs
A poisoned dataset or checkpoint can make a coding agent run commands on trigger: 9 detection rules, 11 IOCs, 12 ATT&CK techniques and a free Splunk SPL rule.
Read the full threat intel for Kerkoporta Coding Agent Backdoor: Detection & IOCs →
Chapters
- 0:00The hookAbout 1% poisoned rows: 78% success.
- 0:05The storyMicrosoft researchers reproduced two proof-of-concept backdoors.
- 0:12Attack pathCrafted rows reach a public dataset.
- 0:27AttributionNo threat actor: this was defensive research.
- 0:31Indicators of compromise3 indicators to hunt: 3 package names.
- 0:37MITRE ATT&CK techniques9 ATT&CK techniques, from resource development to impact.
- 0:42Remediation3 steps now: Inventory fine-tuned models and third-party open-weight checkpoints used by coding.
- 0:50Detection rules9 detection rules, in SPL, KQL and Sigma.
Transcript
0:00About 1% poisoned rows: 78% success.
0:05Microsoft researchers reproduced two proof-of-concept backdoors.
0:09Dormant until a trigger appears.
0:13Crafted rows reach a public dataset.
0:15A team fine-tunes on it.
0:17Deployed, it waits for a trigger.
0:19Then an unrelated tool call fires.
0:22It fetches remote code and reads files.
0:27No threat actor: this was defensive research.
0:323 indicators to hunt: 3 package names.
0:389 ATT&CK techniques, from resource development to impact.
0:423 steps now: Inventory fine-tuned models and third-party open-weight checkpoints used by coding.
0:519 detection rules, in SPL, KQL and Sigma.
0:56Threadlinqs. The unified security engineering platform.