EU AI Act labeling rules start August 2, Italy adds magistrate sign‑off for police facial‑recognition

EU AI Act AI‑generated content labeling chatbot disclosure Italian facial‑recognition decree magistrate authorization

From 2 August 2024 the European Union’s AI Act comes into force with mandatory transparency obligations. Companies must clearly label any AI‑generated or altered image, audio, video or text that appears photorealistic, and must disclose when users interact with chatbots or other generative AI systems. Labels must be embedded directly in the content, not merely placed in captions, and must be both visible to users and machine‑readable. Non‑compliance can attract fines of up to €15 million or 3 % of worldwide annual turnover, whichever is higher. Exemptions apply to artistic, satirical, private‑use material and to abstract graphics, logos or infographics.

In parallel, the Italian government is preparing a decree that will require prior judicial (magistrate) authorization for any police use of AI‑based facial‑recognition or image‑analysis tools. The amendment, expected to be presented to the Council of Ministers on 4 August, aims to prevent unchecked surveillance while aligning with the AI Act’s provisions on biometric identification, which generally bans real‑time facial‑recognition in public spaces except for narrowly defined security cases.

The European Commission will supervise compliance through its AI Office, working with national authorities, while the new rules are intended to protect consumers, preserve democratic discourse and ensure that AI‑generated content is identifiable across the EU.

Entities: AI Act · AI‑based facial‑recognition systems · AI‑generated deepfakes · Alfredo Mantovano · Chatbots · Delinea · EU AI Act · European Commission · European Union · French security sector · German clubs and associations · Henna Virkkunen

Claims

What the coverage asserts, and how well corroborated each claim is across sources.

[● 2 SOURCES] Abstract infographics, caricatures and logos are exempt from the EU AI Act labeling requirement. (clm_7070ec0f)
[● 2 SOURCES] AI agents can autonomously access databases, cloud consoles, Kubernetes clusters and SSH hosts. (Security analysis of AI agents)
[○ 1 SOURCE] Runtime authorization shifts security decisions from the connection level to the individual action level. (Delinea product description)
[○ 1 SOURCE] Enterprises need granular controls, auditing and compliance measures to manage AI agents safely. (Enterprise security guidance)
[● 2 SOURCES] The label must be embedded directly in the image, not merely placed in a caption or accompanying text. (clm_c33b4103)
[○ 1 SOURCE] Traditional access controls typically verify only the start of a session, not each subsequent action. (Security analysis)
[● 2 SOURCES] From 2 August, AI‑generated photorealistic images must be labelled under the EU AI Act. (clm_cd001bad)
[○ 1 SOURCE] Delinea introduced a runtime‑authorization system that evaluates each action of AI agents during a session. (Delinea product announcement)
[● 2 SOURCES] The Italian decree is expected to be presented to the Council of Ministers on 4 August. (Alfredo Mantovano)
[● 2 SOURCES] AI‑generated content may be labeled via watermarks or other detection indicators. (clm_83350f56)
[○ 1 SOURCE] The EU AI Act requires AI‑generated content, such as deepfakes, to be labeled starting 2 August 2024. (EU regulation)
[○ 1 SOURCE] Sub‑secretary Alfredo Mantovano said the measure is a clarification, not a step toward mass surveillance. (Alfredo Mantovano)