Sigma

From October 28, KRiBSI may penalize companies for AI Act violations

2 min read

AI-generated image: From October 28, KRiBSI may penalize companies for AI Act violations AI

Poland's artificial intelligence supervisory authority is gaining powers to inspect and impose penalties. This directly affects companies that use chatbots or AI-generated content in customer service and sales.

From October 28, the Commission for the Development and Safety of Artificial Intelligence (KRiBSI) will be able to inspect companies, conduct proceedings, accept complaints, and impose penalties for violations of the EU AI Act. The Act on artificial intelligence systems itself has been in force since August 11, but until October 28 it is not possible to submit either a complaint or a request for an opinion to the commission, according to "My Company Polska".

The amount of penalties is set by the AI Act in three tiers. For prohibited practices, such as emotion recognition of employees, the penalty may be up to EUR 35 million or 7 percent of global annual turnover. For violating transparency obligations, the penalty may be up to EUR 15 million or 3 percent of turnover, and for misleading the regulator up to EUR 7.5 million or 1 percent. For large companies, the limit is the higher of the two amounts; for small and medium-sized companies, the lower one. These are maximum ceilings, not fixed rates.

The transparency obligation covers chatbots and voicebots (automated systems that conduct voice conversations) in customer service, complaints handling, or banking applications. Under Article 50 of the AI Act, the customer must know that they are speaking with AI, at the latest during the first interaction. In chat, a short message above the conversation field is enough; on a helpline, a voice message before the conversation; a clause in the terms and conditions is not enough, Forsal.pl points out. The obligation rests mainly with the system provider, but the company implementing it should check whether it is being met.

Commission chair Pamela Krzypkowska announced that one of the first inspection topics will be labeling deepfake content (realistic images, recordings, or voice generated by AI that look like real people or events). The first KRiBSI meeting must take place by November 11 at the latest, and a department has been established at the Ministry of Digital Affairs to support the commission and recruit staff. Experts interviewed by the Polish Press Agency (PAP) in August warned that without specialists and tools, autumn inspections and penalties may remain theoretical.

Companies unsure whether their system complies with the regulations may apply for an individual opinion from October 28. The commission has 30 days to issue it, or 60 days in particularly complex cases. According to PAP, the fee will be PLN 150. Employees' mere use of ChatGPT to write emails or notes does not violate the AI Act regulations.

Key takeaways

The summary could not be loaded. Refresh the page and try again.