Nos derniers articles

“... we argue there are good reasons for skepticism, as many of its key operative provisions delegate critical regulatory tasks to AI providers themselves, without adequate oversight or redress mechanisms. Despite its laudable intentions, the AI Act may deliver far less than it promises.”

The increasing complexity of data protection laws, rising compliance costs, and evolving cyber threats make data security a vital business concern.

The current global #dataprivacy situation resembles the accountability crisis during the early 2000s US accounting scandals. Lack of oversight, #transparency, and #regulation has led to confusion and distrust. By emulating successful models like the Sarbanes-Oxley Act, companies can regain consumer trust by treating privacy policies like #financialstatements, standardized and audited. The proposal includes #privacy #controls similar to financial internal controls and a Privacy Cube framework for #riskmanagement, ultimately aiming to rebuild #consumertrust in #data handling.

This is a note on the #gdpr and the use of #us-based #cloudservers. The note raises concerns about the #risk of US #intelligenceagencies having access to #data transferred to any US cloud from the #eu, or directly accessed by US agencies, even while still in the EU / #eea or while in transit. The note discusses cases in #france, the #netherlands, and #germany that have addressed these issues, concluding that the legality of the use of US cloud servers and solutions remains problematic.

This paper presents an extension of #statistical inference for smoothed quantile estimators from finite domains to infinite domains. A new truncation methodology is proposed for discrete loss distributions with infinite domains. #simulation studies using several distributions commonly used in the #insuranceindustry show the effectiveness of the methodology. The authors also propose a flexible bootstrap-based approach and demonstrate its use in computing the conditional five number summary (C5NS) for tail risk and constructing confidence intervals for each of the five quantiles that make up C5NS. Results using #automobile #accident #data show that the smoothed quantile approach produces more accurate classifications of tail #risk and lower coefficients of variation in the estimation of tail #probabilities compared to the linear interpolation approach.

This paper explores the #uncertainty around when #data is considered "#personaldata" under #dataprotection#laws. The authors propose that by focusing on the specific #risks to #fundamentalrights that are caused by #dataprocessing, the question whether data falls under the scope of the #gdpr becomes clearer.

"... nothing meaningful for regulation can be determined solely by looking at the data itself. Data is what data does. Personal data is harmful when its use causes harm or creates a risk of harm. It is not harmful if it is not used in a way to cause harm or risk of harm."

"An employee may be attacked by a potentially sophisticated adversary whose goal is to steal all their data. Therefore, the firm trades off the efficiency benefit of the more permissive data access architecture with the adversarial risk it incurs. We characterize the firm's optimal data access architecture and investigate how it depends both on the adversarial environment and the firm's technology."

"Our evidence also implies that client firms that share the same audit office as breached firms increase their disclosure of cybersecurity risk and their demand for cybersecurity human capital. Reconciling with the Bayesian learning theory, these effects only manifest for auditors located in states that have been only sporadically exposed to data breaches."

"There is currently limited information on and a lack of a unified approach to AI and ESG, and a need for tools for systematically assessing and disclosing the ESG related impacts of AI and data capabilities. I here propose the AI ESG protocol, which is a flexible high-level tool for evaluating and disclosing such impacts..."