Authors: Eoghan Casey, Jim Jones, Jessica Hyde, Cameron Kelley, Chad Steel
DFRWS USA 2026
Abstract
This interactive practitioner-led working session addressed the questions we actually
face when AI enters the digital forensics workflow: When can a model’s output be trusted? How
do error rates and hallucinations translate into conclusions that survive scrutiny? Where is AI
genuinely useful—lead generation, document and email classification, coding assistance—and
where does it have no business driving the decision, such as adjudicating whether multimedia
depicts a child? What are principles of safe and responsible use of AI in digital forensics, and
how can some of these be built in as guardrails? Five practitioners share experience-grounded
perspectives, and live participant polls place the room’s own problems, fears, and judgment at
the fore. Threaded throughout this session is the conviction that the examiner’s responsibility to
critically evaluate and validate results is non-delegable—that we cannot push a button and
blindly accept the output or pass it off to a lawyer or judge to determine whether or not it is
reliable.
Presenters (alphabetical):
● Eoghan Casey — Cyber Knowledge Engineering.
● Jim Jones — George Mason University.
● Jessica Hyde — Hexordia.
● Cameron Kelley — Mobasi.
● Chad Steel — George Mason University.