FRANCE Trends and Developments Contributed by: Eric Weil, Victor Gozzerino and Giada Cancemi, Weil & Associés
Weil & Associés 26 avenue de la Grande Armée 75017 Paris France
Tel: +33 014 415 9898 Fax: +33 014 415 9899 Email: info@weil-paris.fr Web: www.weil-paris.fr
AI’s Double Edge: Accelerating Financial Fraud and Enabling Its Detection AI as an accelerator of financial fraud Artificial intelligence is fundamentally reshaping the scale and sophistication of financial fraud. Established schemes – CEO fraud, spear-phishing, identity theft, phishing campaigns, quishing, ping calls, and deepfakes – are becoming more convinc - ing, faster to deploy and harder to detect. The use of AI in the commission of a wide range of criminal offences has now emerged as one of the most signifi - cant operational risks facing corporates and financial institutions alike. This risk is no longer prospective. The Arup case in Hong Kong demonstrated that a deepfake video-con - ference fraud could lead an employee to authorise multiple wire transfers after believing they were com - municating with genuine company executives, all of whom had been artificially generated. Identifying liability behind the tool At this stage, few French decisions expressly mention artificial intelligence in the context of financial fraud cases. AI does not fundamentally alter the application of the general principles and established categories of French criminal law. It is simply a means by which the physical element of an offence is carried out. Where the relevant statutory provision permits, the constitu - ent elements of the offence will fall to be assessed against the user, to whom criminal responsibility may be attributed. It is therefore a matter of applying, in the traditional manner, the rules on the criminal liability of natural persons or legal entities where the conditions
are met (Criminal Code, Article 121-1; Criminal Code, Article 121-2). The position becomes more nuanced, however, where an AI system is no longer merely executing a simple instruction. Generative AI models such as ChatGPT, Gemini or Claude can produce outputs that were not entirely determined by the user, as they exercise a degree of creative latitude in the means by which they achieve the desired result. This blurs the causal link between the individual, the tool and the harm caused, as well as the requisite mental element needed to establish the offence. While AI may materially contribute to the commis - sion of an offence, it cannot, under French law as it currently stands, be prosecuted as an autonomous author, given that it possesses neither legal personal - ity nor distinct criminal capacity. Article 121-4 of the Criminal Code refers to a “person” who commits or attempts to commit an offence. Given that criminal statutes must be strictly construed, the absence of legal personality prevents criminal liability from being directly attributed to an AI system. It therefore follows that only a legal person (natural or corporate) can bear criminal liability for an offence committed through AI. Developers and users of AI systems are the natural candidates when fault can be established and causally linked to the conduct of the AI that gave rise to the offence. The search for the party who controls the risk also prompts a re-examination of the concept of complic - ity. A user who deploys an AI system will remain the direct perpetrator of the offence. The position is less
64 CHAMBERS.COM
Powered by FlippingBook