AI Gemini

I have recently posted on this blog a lot. As the blog’s manifesto explicitly mentioned the Middle East as a focus, the ongoing war of the United States/Israel with Iran, a country which I had visited nine times between 2003 and 2009, resulted in a new interest with the country and its people. My sentiments are torn. I have met different kind of Iranians, mostly friendly people. Despite that I definitely cannot live there for much longer than the 10 days or two weeks in 2008/2009 when I had been invited for giving a postgraduate course at the Islamic Azad University in Khorasgan.

Yesterday, I browsed through different posts I had written around Easter, when a botched “boots on the ground” operation took place south of Isfahan. I had listened to CBS 60 Minutes’ interview of an Air Force officer called Bravo whose F-15 fighter jet was shot down on 3 April. He was severely injured and miraculously rescued days later by his American fellow soldiers. When Bravo pronounced Iran consistently eye-‘ran (a telltale sign for real haters of the country and its people. In fact it must be ee-‘rahn), I stopped watching.

I remembered that I had reported, in subsequent posts on this blog, that the Wall Street Journal had reported a screaming-for-hours Donald Trump who had to be be moved to another room when the rescue operation took place. I asked two times AI Gemini whether it could confirm my memories, to no avail.

Today, I made another attempt. This time I did it in German (the question is translated into English by GoogleTranslate and slightly edited).

Let me try this again. During the course of the Iran crisis, there were reports of a White House crisis meeting where President Donald Trump apparently shouted for hours and was escorted out of the room. When and on what occasion did that happen? (There is so much reporting about Trump that it’s easy to lose track.)

AI Gemini responded (again, translated by GoogleTranslate).

The incident you are referring to took place in April 2026, when a US F-15 fighter jet was shot down over Iranian territory during the Iran War. [1]

Reports regarding Donald Trump’s angry outburst and his exclusion from the operational crisis meeting are based on a detailed exposé. [1, 2]

Here are the details of the situation at the time:

  • The Trigger: On Good Friday (April 3, 2026), Iran shot down a US fighter jet, leaving two American service members missing. The US military had to plan a high-risk rescue operation in hostile territory under extreme time pressure. [12]
  • Behavior in the White House: Upon learning of the shoot-down and the missing service members, Trump reportedly suffered a massive fit of rage. He allegedly spent hours shouting at advisors in the West Wing and warning—in a panic—of a political disaster. He repeatedly drew parallels to the 1979 Iran hostage crisis, which had cost Jimmy Carter his re-election. He loudly demanded immediate, ill-considered military action. [1234]
  • Exclusion from the Situation Room: Fearing that his extreme impatience and impulsive interference would jeopardize the highly complex rescue mission, military leaders and top advisors made an extraordinary decision: they deliberately kept the President out of the operational planning phase in the Situation Room and isolated him in the West Wing. Instead of providing minute-by-minute updates, they briefed him only by phone at select, critical moments. The mission was coordinated in his absence and ultimately succeeded. [123]

I also replied.

Interesting. I asked the same question in English yesterday. The AI ​​Gemini consistently refused to answer it the way it did today. Are there different loyalties on the two sides of the Atlantic?


I found AI Gemini’s new response so remarkable that it motivated me for this post.

There are no differing loyalties between the versions; however, there are different safety filters and technical mechanisms that apply depending on language and region.

The fact that Gemini blocked the request yesterday but is answering today is primarily due to two key factors:

1. Stricter filters regarding elections and active politicians

Safety guidelines for AI models are extremely sensitive when it comes to current political figures (such as Donald Trump) and sensitive topics (such as wars or the White House).

  • The language barrier: Safety filters are often trained first—and most restrictively—for English, as the majority of global queries are made in that language. Yesterday, the English filter likely misclassified your question as “political disinformation” or “impermissible criticism of an active politician” and refused to provide an answer entirely.
  • The German filter: Filters for other languages ​​(such as German) react differently in certain nuances; they may be less rigid when dealing with historical retrospectives of fictional or real events, provided there is no risk of acute election interference.

2. The “context effect” within the session

An AI model begins every new conversation from a different mathematical starting point.

  • If a question in English is phrased somewhat awkwardly (e.g., using terms like “escorted out” or “screaming”), algorithms designed to prevent hate speech or fake news may trigger immediately.
  • Today, you asked the question in German and provided context with the phrase “one loses track of things” (man verliert den Überblick). This allowed the model to recognize that you were seeking historical, chronological information rather than attempting to engage in political mudslinging.

Conclusion: The AI ​​does not take sides with America or Europe. It is simply subject to a complex, sometimes unpredictable system of filters—one that, in English, is often set so strictly for names like “Trump” that it blocks even legitimate questions regarding media reports.

Some context by Deutsche Welle at that time.

15 September 2026 @ 16:55 UTC+2.

Last Modified September 15, 2026.

This entry was posted in Uncategorized and tagged , , , , . Bookmark the permalink.