Robustness and failure modes
How AI-assisted investigative technology behaves when real evidence departs from ideal benchmark assumptions, and how those limits should be communicated.
Research portfolioResearch theme Research theme Research theme Planned direction Planned direction Planned direction Planned direction Technical article · Red Hot Cyber
Trustworthy AI for digital investigations requires visible limits, not only good outputs.
My research interests focus on a practical question: when AI is used in a forensic or investigative workflow, how can its output be relied upon, challenged and documented? The emphasis is on robustness, traceability, provenance, privacy, human validation and technical accountability.
Current research themes
Public descriptions are intentionally kept at the level of the research problem rather than disclosing details of work currently subject to anonymous scholarly review.
Human validation
Where automated assistance can support analysis and where qualified human review must remain decisive in evidence-sensitive work.
Technical accountability
How model version, input boundary, uncertainty, provenance, tool effects and human responsibility can be made auditable.
Research trajectory
The broader agenda concerns trustworthy AI for digital forensics and investigation. The areas below are forward research directions rather than claims of completed results.
Agentic AI for forensics
How autonomous or semi-autonomous AI agents can assist investigations while preserving human authority, evidence boundaries, provenance, auditability and explicit approval points.
Anti-AI forensics & authenticity
Methods for assessing whether multimedia evidence is authentic, manipulated or AI-generated, including provenance signals and uncertainty-aware conclusions.
Private / local forensic AI
On-premise and local-model architectures for sensitive forensic data where confidentiality, data control and reproducibility matter more than cloud convenience.
Technical governance
Operational controls for documenting model use, evidence boundaries, uncertainty, validation and human decision responsibility.
Research principles
Measurable
Claims should be tied to observable behaviour, explicit evaluation criteria and documented limitations.
Reproducible
Methods should preserve traceability of inputs, assumptions, tool conditions and decision points.
Operationally relevant
Research should explain what a failure means for evidence handling, investigation and human decision-making.
Publications & technical writing
Research themes are complemented by technical writing on operational digital forensics and evidence-sensitive decision-making.
Digital Forensics: acquisire subito o aspettare? Prima bisogna capire cosa rischiamo di perdere
First-response choices around live systems, volatile evidence, encryption and acquisition are treated as state-dependent decisions with explicit consequences for preservation and evidential value.
Positioning: the central theme is not “AI applied to forensics” in the generic sense. It is whether AI-assisted investigative technology can be made robust, verifiable, privacy-preserving and technically accountable enough for evidence-sensitive work.