Menu
SUBSCRIBE
SUBSCRIBE
Menu

AI-Generated Content

European Union flags outside an EU building for coverage of the AI-generated content code.

190 Organizations Back EU Code for Labeling AI-Generated Content

, , , , , , , ,

The voluntary code provides an approved way to meet binding EU requirements for marking AI output and labeling deepfakes and certain public-interest text.

The European Commission’s live signatory list now names 190 organizations that have backed its voluntary Code of Practice on Transparency of AI-generated Content.

The code gives AI providers and companies using generative AI an approved way to show how they meet European Union requirements for marking and labeling AI-generated content.

It divides responsibility between providers, which develop and offer AI systems, and deployers, which use those systems for business or professional purposes. Organizations can sign the section that applies to them or both.

AI providers must make outputs detectable

Providers must place machine-readable markers in AI-generated or manipulated text, images, audio, and video. These digital markers must allow automated tools to identify the material as AI-generated content.

The requirement applies to providers offering AI systems in the European Union. Providers based elsewhere are also covered when their systems’ output is used within the bloc.

The code gives signatories a common set of measures they can use to demonstrate that their marking and detection methods comply with the AI Act.

Companies publishing AI content face separate labels

Organizations using generative AI must clearly label deepfakes and certain AI-generated or manipulated text.

The deepfake requirement covers images, audio, and video that resemble real people, objects, places, organizations, or events and could appear authentic.

Companies must also label AI-generated text on matters of public interest, including politics, public health, finance, science, and culture. No label is required if the text receives substantive human review and someone accepts responsibility for publishing it.

A spelling or grammar check does not qualify as human review. The Commission also said a machine-readable marker added by the AI provider does not replace the clear label required from the company publishing the content.

Signatories span technology and other industries

The provider section currently has 82 signatories, while the section for companies using AI has 152. Some organizations signed both, leaving 190 distinct names on the Commission’s list.

Anthropic, Google, Meta, Microsoft, and OpenAI are among the companies backing the provider requirements. Bulgari, Getty Images, Lenovo, and Lufthansa are among those supporting the rules for organizations using and publishing AI-generated content.

Signatories also come from the information technology, telecommunications, education, and retail sectors. The Commission described about half of the signatories as small, recently established companies.

The code is voluntary, but EU requirements are not

Signing the code is voluntary. The underlying AI Act transparency requirements became legally binding on August 2.

Signatories can rely on the code to demonstrate compliance across EU member states. Organizations that do not sign must show compliance through other adequate measures and may receive more requests for information from regulators.

AI systems placed on the market before August 2 have until December 2 to meet the machine-readable marking requirement. The grace period does not apply to the other transparency requirements.

National regulators will handle most enforcement. Violations can bring fines of up to €15 million or 3% of worldwide revenue from the preceding financial year.

In September, the European Union’s AI Office plans to launch two task forces and invite signatories to participate. Members will compare how they are applying the transparency measures and explore better ways to put the code into practice.

Five candidate lawn signs on a suburban street illustrate election fraud targeted by the AI Ads Act.

AI Ads Act Targets Fraudulent AI Election Content

, , , , , , , , ,

The bill would write AI-assisted fundraising fraud into the Federal Election Campaign Act and expand impersonation rules beyond candidates and their representatives.

 

The AI Ads Act would explicitly put AI-generated election content under Section 322 of the Federal Election Campaign Act, which prohibits falsely claiming to represent a candidate or political group, including to solicit donations.

Rep. Ro Khanna (D-Calif.) introduced H.R. 9985. The House referred the bill to the Committee on House Administration. Sen. Adam Schiff (D-Calif.) introduced a Senate version.

The bill would not ban all AI use in political advertising. The content would have to fraudulently present a person or group as speaking, writing, or otherwise acting for a candidate, political party, committee, or political organization.

AI fundraising fraud would fall under the law

Federal law already prohibits a person from pretending to speak, write, or otherwise act for a candidate or political party, or their employees or agents, to solicit political contributions or donations.

H.R. 9985 would expressly state that this prohibition covers content generated in whole or in part with AI. It would also extend the rule to fraudulent solicitations made in the name of a real or fictitious political committee or organization.

The bill would not create a separate requirement to label AI-generated fundraising material. Instead, it would expand the existing federal ban on fraudulent political solicitations.

Bill would write AI into existing impersonation law

Section 322 of the Federal Election Campaign Act already prohibits two forms of fraudulent impersonation. Candidates and their representatives cannot pretend to act for another candidate or political party to damage them. The law also prohibits anyone from pretending to represent a candidate or political party to solicit contributions or donations.

The Federal Election Commission (FEC) said in 2024 that these restrictions apply whether the fraudulent content is created with AI or another technology. However, that guidance only explained how the FEC interprets the existing law. It did not change the statute, which does not mention AI.

H.R. 9985 would amend Section 322 to state directly that either violation can involve content generated “in whole or in part” with AI. This would put the FEC’s interpretation into federal law instead of leaving it only in agency guidance.

Using AI would not create a violation by itself. The content would still have to falsely present someone as speaking or acting for a candidate or political group.

Impersonation rules would cover more people and groups

The bill’s broader change would be to expand who can violate the law and who can be impersonated.

Current law generally limits the campaign-impersonation provision to candidates and their employees or representatives. It also requires the impersonation to damage another candidate or political party.

The AI Ads Act would remove both restrictions. Any person could be held responsible for falsely claiming to represent a candidate or political group, and the government would not have to prove that the impersonation was intended to damage someone.

The bill would also extend both the impersonation and fundraising provisions to false claims of representing real or fictitious political committees and organizations.

Existing enforcement and penalties would apply

The bill would not establish a separate penalty for using AI. Violations would be handled under the existing federal campaign-finance enforcement system.

Anyone could file a formal complaint with the FEC. If at least four FEC commissioners agree, the agency could investigate and try to resolve the case through corrective action and a civil penalty. If no agreement is reached, the FEC could take the case to court. It’s important to note that the FEC generally has six commissioners, but presently it only has two. With four seats vacant, the FEC could not act on complaints.

Knowing and willful violations can also result in criminal prosecution, with fines or imprisonment available under existing federal law.

There would be no automatic penalty simply because political content was created with AI. Authorities would still have to establish that the person fraudulently claimed to speak or act for a candidate or covered political group.

H.R. 9985 remains before the House Administration Committee.

European Parliament building and national flags illustrate EU AI transparency rules taking effect August 2.

EU Details AI Transparency Rules Taking Effect August 2

, , , , , , , , ,

The rules divide responsibility between companies that provide AI systems and those that use them, with maximum fines set at the higher of €15 million or 3% of worldwide annual revenue.

 

The European Commission published guidelines Monday explaining how companies must follow EU AI transparency rules that take effect Aug. 2.

The rules require companies to disclose when people are interacting with AI or being analyzed by systems that assess emotions or use biometric data to categorize them. They also require companies to make AI-generated content detectable and clearly label deepfakes. 

The Commission issued the guidelines less than two weeks before Article 50 of the EU AI Act takes effect. The 50-page document explains which systems and content fall under the rules, what companies must disclose and how regulators will enforce the requirements.

Providers must build transparency into AI systems

Under the AI Act, a provider is a company or organization that develops an AI system itself, or hires another party to develop it. Then it makes the system available or begins using it under its own name. This includes systems sold to customers, provided for free or developed for the company’s own internal use. 

Providers of AI systems that interact directly with people must make clear from the first interaction that the person is dealing with AI. The disclosure is not required when the AI interaction is already obvious, although the guidelines say regulators should apply that exception narrowly.

Providers of generative AI systems must also place machine-readable marks in synthetic text, images, audio and video. Those marks must allow automated tools to detect that AI generated or changed the content.

Standard editing that does not substantially change the original material is generally outside that requirement. The guidelines also exclude source code, short strings of numbers or symbols, and content used only in closed industrial or production systems before it becomes a final output.

Businesses using AI have separate duties

The AI Act defines a deployer as a company or organization using an AI system under its authority. Employees and contractors operating the system under the company’s control are not treated as separate deployers.

Deployers using emotion recognition or biometric categorization must tell people exposed to those systems.

Companies must also clearly label AI-generated or altered images (i.e., “Deepfakes”) and audio or video that could falsely appear authentic. The disclosure must be visible or audible by the time a person first encounters the content. A hidden machine-readable mark alone is not enough.

AI-generated or altered text published to inform the public about matters such as politics, public health, consumer safety, or financial developments must also carry a label. The label is not required when the text has received either substantive review by someone with relevant knowledge or meaningful editorial review, and a person or organization accepts legal responsibility for its publication. Spell-checking or grammar correction alone does not count. 

Deadline, grace period and fines

Most EU AI transparency rules apply starting Aug. 2. The guidelines provide a limited grace period extending to Dec. 2 for systems placed on the market before Aug. 2, but only for the requirement to add machine-readable marks to synthetic content. Content generated before Aug. 2 does not have to be labeled retroactively. 

Government regulators appointed by each EU country will investigate most violations and impose penalties. The Commission’s AI Office will handle cases involving systems built on general-purpose AI models when the same company provides both the model and the finished system. It will also oversee AI systems integrated into the EU’s largest online platforms and search engines. The European Data Protection Supervisor will enforce the rules for AI used by EU institutions and agencies. 

For companies, fines can reach the higher of €15 million or 3% of worldwide revenue from the previous financial year. For small and medium-sized companies, the maximum is the lower of those two amounts.

The guidelines explain how the Commission interprets the AI Act but are not legally binding. The EU has separately issued a voluntary Code of Practice that sets out specific methods for marking and labeling AI-generated content. Companies can formally commit to following the code by having a senior executive sign and submit the required commitments to the European Commission. Companies that follow those commitments can point to them as evidence of compliance. Those that do not join must show regulators that their own methods meet the law equally well.