Talos studies prompt logs as a new class of forensic artifact

Findings and evidence

Cisco Talos examines a corpus of artifacts left by adversarial AI use, including local prompt logs. The researchers identify several categories of observed activity and argue that the operator’s existing skill remains relevant to the outcome.

Why it matters

Where authorized and relevant, preserve assistant conversation logs with their surrounding filesystem and process context. Treat generated claims as statements within evidence, not as proof that the claimed actions occurred.

Scope and limits

The collected corpus is not a random sample of all AI users or attacks. A prompt log can show intent or discussion without proving successful execution.

Primary source

Cisco Talos: original publication. Source published 2026-08-04. Brief prepared by websec.gr on 26 September 2026. This is an editorial research summary, not a claim of independent replication.