The EU prohibits the use of AI-driven virtual assistants during online meetings. Such assistants are often used to transcribe, note or even detect visuals and audio during a video conference.
In a presentation from the Europe Commission delivered to European digital innovation nodes earlier this month, there is a note about “online meeting ethics” -glass saying “No AI agents are allowed.”
AI Agents are tools that can perform complex, multi-step tasks autonomously often by interacting with applications, such as video conferencing software. For example, Salesforce uses AI agents to call sales managers.
The Commission confirmed that this presentation was the first time this rule was introduced but refused to explain why Politico was asked. There is no specific EU legislation covering AI agents, but the AI ​​models that drive them will have to comply with the strict and controversial rules of the AI ​​law.
AI agents raising security concerns
While AI Notetakers and other agent types are not in themselves a security threat, according to a 2025 report from Global AI experts, security risks come from the user who is not aware of what their AI agents are doing, their innate ability to operate outside the user’s control and potential AI-AI-interactions. These factors make AI agents less predictable than standard models.
SEE: How can AI be used safely? Researchers from Harvard, MIT, IBM & Microsoft weigh in
Technical companies must be careful when promoting products that can achieve an increasing amount without the user’s attention. One of the biggest precautions is Microsoft Recall, an AI tool that allowed users to control their PC or search through files using natural language.
The convenience came at a price: Remember caught screening of active windows every few seconds, saving them as a timeline, raising concern for privacy and data and led to significant launch delays. Microsoft has since released a number of agents specifically designed to tackle cyber threats.
AI agents are growing in propagation
This has not prevented the AI ​​players from handing over more control to their models. Anthropic added a computer use feature to its Claude Sonnet -Chatbot in October 2024, which gave it the opportunity to navigate desktop apps, move markers, click buttons and write text. Its deep research function, which was announced this week, also responds to requests “agentic”, just as Microsoft’s equivalent does.
Last month, Openai expanded its text-to-speech and speech-to-text tools for agent models, indicating their growing relevance. In January 2025, Openai Operator announced, an agent tool that runs in browser to autonomously perform actions such as ordering groceries or booking trips.
SEE: The EU invests $ 1.3 billion. EUR to increase AI -adoption and improve ‘digital skills’
Anthropic and Openai even work together to improve agent technology, with the latter adding support to the former Model context protocol, an open source standard to connect AI apps, including agents, to data stores. Anthropic has also joined forces with databicks to help large business customers build their own agents.
TechPublic predicted by the end of 2024 that the use of AI agents will increase this year. Openai -Managing Director Sam Altman repeated this in a blog post in January and said “We can see the first AI agents ‘participate in the workforce’ and change the material’s output material.”
By 2028, 33% of Enterprise software applications will include Agentic AI, up from less than 1% by 2024, according to Gartner. One fifth of online store interactions and at least 15% of daily work decisions will be held by agents of that year.