Intelligent Security Inspection of Data using forensic Extraction and Analysis

map[class:object-cover object-center src:images/home/hero/fis2.png]

Digital Forensics

Technische Hochschule Mannheim

Artificial Intelligence

University of Twente, ITIS GmbH

Legal Assessment

CII, FZI

Project Management

ISS

map[class:aspect-square object-contain object-center w-[75%] src:images/home/features/features-1.png]

Digital Forensics

Chain of Custody, Digital Evidence

Digital Forensics of AI

In this modern age, crime most of the time has also a digital dimension. Criminals are using technology to facilitate their offenses and avoid apprehension, creating new challenges for attorneys, judges, law enforcement agents, forensic examiners, and corporate security professionals. One of the key technology is Artificial intelligence (AI). To ensure reliability and to audit internal decisions Digital Forensic Readiness of existing and new AI systems needs to be considered.

map[class:aspect-square object-contain object-center w-[75%] src:images/home/features/features-2.png]

Artificial Intelligence

Text-based AI

Complex and self-learning AI models

The forensics of text-based AI systems face particular challenges because the traceability of model-internal decision-making processes is limited and manipulations are difficult to prove conclusively. At the same time, attackers often use subtly worded inputs that are almost indistinguishable from regular requests and thus make forensic classification as an attack considerably more difficult. At the same time, attackers often use subtly formulated inputs that are difficult to distinguish from regular queries, making forensic classification as an attack considerably more difficult. In addition, specialized attack techniques such as inversion, in which attackers attempt to reconstruct confidential training data from the model, or evasion, in which inputs are deliberately manipulated so that the system classifies them as incorrect or harmless, make it difficult to clearly identify malicious interactions. The analysis becomes particularly complex in continuously learning systems, whose constantly changing state limits the reproducibility of results. When models are dynamically updated, it is often not possible to log all interactions completely, which makes it even more difficult to reconstruct possible attack paths and clearly assign security-related incidents.

map[class:aspect-square object-contain object-center w-[75%] src:images/home/features/features-3.png]

Legal Assessment

Attorneys, judges and law enforcement agents

Challenges of using AI for attorneys, judges, law enforcement agents

To ensure that forensic investigations are legally valid in court, legal support is necessary. This support covers a wide range of regulations, including European regulations such as the AI Act and GDPR, product law, protection of trade secrets and national procedural law. The purpose of this support is to analyse the requirements for AI applications and their impact on judicial usability, based on formal and substantive law, and to identify regulatory gaps. This helps to reconcile the social interests associated with the use of artificial intelligence. The responsibility for this lies with the Cyberintelligence Institute (CII) and the FZI Research Center for Information Technology.

Project Partners

map[alt:Technische Hochschule Mannheim class:h-16 w-auto src:images/home/clients/thm.png] map[alt:University of Twente class:h-16 w-auto src:images/home/clients/ut.png] map[alt:ITIS GmbH class:h-16 w-auto src:images/home/clients/itis.png] map[alt:CII class:h-16 w-auto src:images/home/clients/cii.png] map[alt:FZI class:h-16 w-auto src:images/home/clients/fzi.png] map[alt:Institute for Security and Safety GmbH class:h-16 w-auto src:images/home/clients/iss.png]