AI Office can request information, evaluate models, and request risk mitigation measures from GPAI suppliers.

The European Commission can apply the obligations of the AI Act for providers of general-purpose artificial intelligence models starting from August 2, 2026, and the AI Office can request information, conduct assessments of the models, and require providers to take measures to mitigate risks. The Commission invoked the new powers at the briefing on August 11, in response to concerns about manipulation risks through bots, fake accounts, and content generated by artificial intelligence.

In short, from August 2, 2026, the Commission can effectively enforce the obligations of the AI Act for providers of general-purpose artificial intelligence models, including through sanctions. The AI Office can request information from providers, conduct assessments of GPAI models, and require remedial measures, including risk reduction. The obligations for GPAI model providers began to apply as early as August 2, 2025. The date of August 2, 2026, marks the enforcement powers of the Commission coming into effect, not the first appearance of those obligations. From August 2, 2026, separate transparency rules will also apply to certain AI systems and certain forms of AI-generated or manipulated content. The Commission presented these tools in response to a question regarding election manipulation and public consultations but did not claim that the new rules could completely eliminate such risks.

The new enforcement powers represent a distinct stage in the implementation of the AI Act. The rules for providers of general-purpose artificial intelligence models, known as GPAI, began to apply on August 2, 2025, but the first year was primarily designed as a period for implementation and cooperation with providers.

From August 2, 2026, the Commission can move to the effective enforcement of these obligations and can utilize the oversight powers provided for GPAI models.

The AI Office is the structure of the Commission specifically responsible for overseeing general-purpose artificial intelligence models. Among its powers are the ability to request information from providers and assess models when necessary to verify compliance with the rules.

The Office can also request measures from providers. In the case of models that present systemic risks, these may include technical or organizational measures to reduce identified risks.

In situations provided for by the AI Act, the Commission can also request the withdrawal of a model from the market and impose fines for non-compliance with obligations.

GPAI model providers must maintain technical documentation about their models and make it available to the AI Office when requested. They must also provide relevant information to developers building AI systems based on these models.

The rules also include obligations regarding copyright compliance and publishing a sufficiently detailed summary of the content used to train the model.

GPAI models considered to have systemic risk are subject to additional requirements. Providers must assess and mitigate systemic risks, monitor and report serious incidents, and maintain an adequate level of cybersecurity.

The enforcement powers coming into effect in August allow the Commission to more directly verify whether these obligations are being met.

At the briefing on August 11, the Commission presented the new powers of the AI Office after being asked whether citizens can consider that there are sufficient guarantees against election manipulation through bots, fake accounts, avatars, and other AI-based tools.

The question also addressed the possibility that automated systems could influence opinion polls or manipulate public consultations of European institutions.

The Commission did not announce an investigation into a specific case in response, nor did it state that it had identified manipulation of a public consultation through artificial intelligence.

The spokesperson indicated, instead, the tools that have become available or applicable since August 2.

Alongside the oversight powers over GPAI models, new transparency obligations for certain artificial intelligence systems began to apply from the same date.

These rules aim for people to be able to recognize certain situations when interacting with an AI system or when exposed to content generated or manipulated artificially.

Providers of AI systems must ensure that users are informed when they interact directly with artificial intelligence.

For certain generative systems, the AI Act also provides for the use of markings that can be automatically detected for content generated or manipulated artificially.

There are also obligations for those who professionally use certain AI systems. These relate, among other things, to informing the public when exposed to deepfakes or, under certain conditions, to AI-generated or manipulated texts published to inform the public on matters of public interest.

The transparency rules and the GPAI regime are two different components of the AI Act.

The AI Office's powers regarding GPAI models focus on providers and the risks associated with general-purpose models. The transparency obligations primarily aim for people to know when they interact with certain AI systems or when certain materials have been produced or modified artificially.

The fact that both stages have an important milestone on August 2, 2026, does not mean that they are the same obligation or that they are applied in all cases by the same authority.

For the transparency rules, oversight primarily falls to national market surveillance authorities. The AI Office intervenes for systems that fall under its jurisdiction, while the European Data Protection Authority has the corresponding role when EU institutions are providers or users of those systems.

The Commission believes that these obligations can reduce the risk of misleading users, as they receive more information about the nature of the system or the content they interact with.

At the briefing, the spokesperson linked these requirements to the possibility that citizens can recognize interactions with artificial intelligence and content generated or modified by AI.

However, the Commission did not state that labeling content or the new powers of the AI Office can prevent any attempt at election manipulation.

The AI Act provides tools for overseeing models and for the transparency of certain uses, while other risks related to platforms, influence campaigns, misinformation, or electoral processes may also fall under other European and national rules.

The new powers allow for the transition from the initial implementation period of the GPAI rules to a stage where providers can be investigated and sanctioned if they do not comply with applicable obligations.

The AI Act came into force in 2024, and its various obligations are being applied gradually. For providers of general-purpose artificial intelligence models, the relevant requirements apply from August 2, 2025.

On August 2, 2026, the Commission's powers to ensure compliance with these obligations became applicable, including through requests for information, assessments of models, measures required from providers, and sanctions.

On the same day, transparency obligations for certain systems and types of AI content also began to apply.

The Commission's statements on August 11 do not announce a new set of rules adopted that day. They clarify the tools that became effectively applicable nine days earlier, in response to risks of manipulation and misleading through artificial intelligence.