EU AI Act mandates transparency for deepfakes and AI content

EU AI Act deepfake regulation AI transparency Spotify AI Persona Anthropic watermarking

New transparency regulations under the EU AI Act have entered into force, mandating the labeling of deepfakes and AI-generated content to prevent public deception. As of August 2, 2026, entities must explicitly signal when users are interacting with chatbots or viewing realistic synthetic media. While visible labeling is required immediately, a deadline of December 2, 2026, has been set for implementing machine-readable technical markings.

In response to these legal shifts, technology companies are implementing various safeguards. Anthropic has introduced an invisible watermark for Claude-generated text designed to persist even after copying or reformatting. Meanwhile, Spotify is addressing the influx of synthetic music by introducing 'AI Persona' labels. Artists can self-identify as AI personas, while the platform will use moderation and algorithms to apply 'Likely AI Persona' labels to non-human accounts.

In Poland, the government is establishing the Commission for the Development and Security of Artificial Intelligence (KRiBSI) to oversee the AI market. This includes the launch of a 35 million PLN AI sandbox project, providing a secure environment for companies to test high-risk systems for compliance with the AI Act. Cybersecurity experts warn that despite these regulations, deepfake fraud remains a significant threat, with reports indicating that nearly 10% of surveyed individuals have encountered such fraud on work equipment.

Entities

AI Act · Amdocs · Anthropic · Autodesk · Cisco · Commission for the Development and Security of Artificial Intelligence · European Union · KRiBSI · Ministerstwo Cyfryzacji · Ministry of Digital Affairs · Poland · ServiceNow

Claims

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

[○ 1 SOURCE] Anthropic's watermarking system for Claude is designed to survive copying and reformatting. www.gry-online.pl
[○ 1 SOURCE] Nearly 10% of surveyed individuals have encountered deepfake fraud on work equipment. next.gazeta.pl
[● 5 SOURCES] The EU AI Act requires the labeling of AI-generated content. www.lemotif.fr · www.gry-online.pl · dorzeczy.pl · dobrewiadomosci.net.pl · dev.to
[● 2 SOURCES] Spotify will apply 'Likely AI Persona' labels to accounts identified as non-human via moderation and algorithms. conowego.pl · spidersweb.pl
[● 3 SOURCES] Visible labeling of deepfakes is required immediately as of August 2, 2026. www.lemotif.fr · dobrewiadomosci.net.pl · dorzeczy.pl
[● 2 SOURCES] KRiBSI will serve as the AI market supervisor in Poland. biznes.interia.pl · brandsit.pl
[● 2 SOURCES] The Polish AI sandbox project has a budget of 35 million PLN. brandsit.pl · biznes.interia.pl