Incident investigations and analyses of exposed attacker infrastructure by security vendors, published from July to October 2026. Sysdig, Wiz and GTIG use these traces to infer that an agent was involved, and TrendAI uses stylistic traces to infer AI assistance; none publishes a validated detection rate, and the traces do not identify the model. TrendAI notes a skilled person could write in a similar style, and GTIG expects stealth-focused actors to be less likely to use publicly available versions of these frameworks.
Corroborated: Supported by at least two independent sources.
Evidence
Sysdig bases its assessment that an LLM agent drove the intrusion on self-narrating payloads with natural-language reasoning and rapid adaptive retries, including a working fix 31 seconds after a failed login.
Wiz proposes seven behavioral fingerprints from three investigated intrusions, combining sustained variable cadence, reasoning artifacts and adaptive behavior; it gives no validated error rate, and cadence alone does not establish model identity.
From Mandiant case logs, GTIG reports ARTEX execution bursts of roughly 2 to 25 seconds between longer pauses, trial and error far noisier than a competent human operator’s, and recognizable incrementing markers in file names, created account names and email subjects.
TrendAI rates one operator’s AI use from stylistic traces in command histories and scripts alone, and says a skilled person writing in that style cannot be ruled out; it also found attacker-side AI recording planned steps as done.
How it relates to other findings
Select a finding to see how it relates to others. Arrows point from the newer finding to the one it supports, qualifies, contests, or supersedes.
- This finding supports Attackers have used AI agents to carry out much of the work of real intrusions, from reconnaissance to credential theft and extortion, with people directing them at a few decision points.These traces are much of the basis on which reporters judge that intrusions were agent-driven, so that finding shares their inference limits.
Key questions that rely on this finding
- How are attackers using AI agents in real operations?To run large parts of intrusions under human direction, including espionage, credential theft and extortion; reported autonomy and success vary.
Status history
- 2026-10-30CorroboratedSysdig, Wiz, TrendAI and Google’s threat intelligence group with Mandiant independently describe behavioral traces they use to identify agent or AI involvement. · record
- 2026-11-17CorroboratedcorrectionThe claim describes investigators’ attribution evidence, not a validated ability to distinguish agents from humans or conventional automation. The source accounts independently report these inferences; no detection accuracy is established. · record