ChatGPT improved on fake election images but Gemini did not

US midterm voting starts on Tuesday. OpenAI and Anthropic set up war rooms to monitor their platforms for misuse, Politico and Semafor report. Meta says that it opened its Election Operations Center in the week before. A 22 October report from the Brennan Center found that ChatGPT improved since May. Gemini and Meta AI agree to make realistic fake election images more easily.

How it works

The war rooms are teams that monitor for misuse and give reliable voting information. Meta says that its Election Operations Center monitors its platforms. Microsoft says that it improved the answers of Copilot to election questions. It also says that malicious deepfakes of candidates are not permitted.

The numbers

The Brennan Center tested six AI image tools. If a AI prompt is dangerous, ChatGPT gives a “cartoonish image” that no person can read as a photograph. Gemini and Meta AI agree to make convincing fake images more easily. The other tools show a small change.

What is not known

OpenAI and Anthropic gave no comment, and the war room information comes only from two reports. Google and Perplexity gave no answer. No source gives test results for the war rooms.

Insight: AI firms build a public record of safeguards before blame comes

Axios reports that OpenAI plans a public relations campaign for after the vote. Its chief global affairs officer calls this election “a movie trailer” for 2028. Executives at OpenAI and Anthropic think that the public and governments will blame AI firms for a large incident. This reads as a move to show safeguards first, and the Brennan Center test gives a public score. Buyers can tell vendors to give the same proof.