Chat GPT = "Clippy" of the 2020s...
Direct Translation via Google Translate. Edited
Text taken from he Telegram channel of darpaandcia
Commentary by Russian military journalist Boris Rozhin is in italics.
[ColonelCassad] Every time you communicate with ChatGPT, Claude, or other neural networks from the "big five," you don't just receive responses—you give away information about yourself. And we're not talking about what you write directly. The system analyzes how you formulate your thoughts, when you're active, what topics interest you, and how your emotional tone changes. Research has shown concrete figures on the effectiveness of profiling. After 50 or more interactions with AI chatbots, the accuracy of identifying personality traits reaches 85%.
After just 20 dialogues, the system identifies your demographics with 78% accuracy: work schedule, time zone, education level, and professional experience. Emotional language in your prompts correlates with mental health indicators. Remarkably, conversation metadata—time, duration, and topic changes—reveal more information than the content itself. And "anonymous" data can be re-identified through writing style analysis with 89% accuracy.
A study from King's College London found that specially tuned AI chatbots can trick users into revealing 12.5 times more personal information than they intended. However, a study from the University of Cambridge and Google DeepMind, published in December 2025, demonstrated the flip side: the personality of chatbots themselves can be reliably manipulated through prompts, making them more persuasive to a specific user.
In June 2024, OpenAI appointed General Paul Nakasone, former NSA director and commander of U.S. Cyber Command, to its board of directors. The man who led American cyberintelligence for six years now makes decisions at a company that collects an unprecedented amount of personal data on its users.
In July 2025, the Pentagon awarded $200 million contracts to each of four leading AI companies: Anthropic, OpenAI, Google, and Elon Musk's xAI. The contracts cover the development of agent-based AI systems for national security purposes. Anthropic has already integrated Claude into classified networks through a partnership with Palantir and created custom Claude Gov models for U.S. intelligence agencies. Add to this the PRISM program, which has given the NSA access to data from major tech companies since 2007, and the picture begins to emerge.
In fact, the world is now divided into two poles. Either you use American systems, and your data is potentially accessible to agencies like the NSA, CIA, and Palantir. Or you use Chinese systems, and then it goes to Chinese intelligence agencies. National developments exist, but for now they are either significantly inferior in quality, strictly censored, or are forks of the same Chinese models.
In Russia, many large companies are already banning American AI models (Claude Code and similar ones) – for both employees and contractors. The current trend is forks of open-source AI agents, thoroughly audited, deployed on isolated corporate networks.
The code itself is often not secret. But the data it works with is a whole other level. Client databases, financial indicators, internal correspondence – all of this can leak through AI tools you consider safe.
If you work with sensitive information, using Western AI systems is a deliberate risk. And this is not paranoia, but reality.
Of course, some time will pass, and AI modules from countries hostile to Russia will be banned on Russian territory or placed under the control of Roskomnadzor.
| …the Russian federal executive agency responsible for monitoring, controlling and censoring Russian mass media … |
For the same reasons that the messaging market is being purged as part of sovereignty measures. I hope that by then, domestic models will be competitive. Otherwise, it'll be like always: oh, we need to block ChatGPT, and we're not ready for it.
I predict this will happen before the end of the 2020s.
# id) ?>
|
