EU AI Act Gets Specific on Election Tech
As the EU AI Act enters its operational phase, new details are emerging for political tech. The Act classifies AI systems in elections as “high-risk,” requiring strict transparency and audit trails. Meanwhile, policymakers are pushing for mandatory watermarking and cross-platform monitoring of all AI-generated political content for the 2026 election cycle.
The "high-risk" classification under the EU AI Act specifically applies to AI systems intended to influence the outcome of an election or the voting behavior of individuals. This category does not, however, include tools used for the administrative and logistical organization of political campaigns, where individuals are not directly exposed to the AI's output. The core concern is to mitigate risks of undue interference with the right to vote and to protect against adverse effects on democracy. Providers of these high-risk systems face stringent obligations that will become effective from August 2, 2026. These mandates include establishing robust risk and quality management systems, ensuring high-quality data governance to prevent biases, and maintaining detailed technical documentation. They must also implement a post-market monitoring system to track performance and report any serious incidents. For deployers, the rules require maintaining automatically generated event logs, ensuring human oversight, and informing any affected individuals that they are subject to the use of a high-risk AI system. The required audit trails are extensive, mandating the logging of each system transaction, the specific data inputs and corresponding outputs, and any instances of human interaction or overrides to make the AI's operation fully reconstructable. Non-compliance carries substantial financial penalties, with fines for violations related to high-risk systems reaching up to €15 million or 3% of a company's total worldwide annual turnover, whichever is higher. Providing incorrect information to authorities can result in fines of up to €7.5 million or 1% of global turnover. The push for watermarking is part of the Act's broader transparency obligations, which also take effect in August 2026. These rules require that AI-generated content, especially "deep fakes," must be clearly identifiable. To facilitate compliance, the EU's AI Office is developing a voluntary Code of Practice, expected to be finalized by mid-2026, which will outline standards for labeling and metadata. Some policy advocates argue that transparency measures like watermarking are insufficient for electoral deepfakes, as they can be easily removed. They propose reclassifying AI-generated electoral deepfakes from "limited risk" to an "unacceptable risk," which would lead to an outright ban on their use. Prohibited AI practices, such as those using manipulative techniques that can cause significant harm, face the highest penalties of up to €35 million or 7% of global turnover.
