Neural networks in the elections Americans found the perfect consultant on election day - and it scared the hell out of the creators of neural networks. Voters are entering more and more queries into search engines and asking chatbots en masse where to go to vote and how to fill out ballots. As a result, giants such as OpenAI and Anthropic have for the first time deployed 24/7 "status staffs" to put out live fires. The reason for concern is very real: recent studies have shown that almost a third of the answers of popular models about the elections contain inconsistencies, outdated addresses or outright errors. How Big Tech Saves Itself From Scandals: Safe Mode Activated. OpenAI and Anthropic agreed with the AP agency and the democracy platform works: instead of free reasoning, chatbots simply send the user through links to official government portals. Toxic subjects were silenced. Google was so confident that its chatbot simply refused to answer a question about the winner of the 2020 election, citing technical difficulties in verifying data. Turn the deep fakes into cartoons. To prevent the spread of fakes about overcrowding in polling stations or explosions using image generators, ChatGPT has included a filter - for each controversial request, the system now generates deliberately unrealistic images. Hundreds of people were manually banned. Companies have dedicated entire departments of researchers who around the clock identify and block networks of bots trying to pretend to be members of election committees. There is also pressure from Washington: influential senators have already sent warnings to the management of the IT companies, directly implying that they will be held responsible for any mistakes or manipulations in the congressional debates. Apparently, the era of free answers has come to an end: trying not to anger politicians, developers are entering more and more strict filters, which is why the smart assistants during elections risk becoming an ordinary board with official links. the alliance