A new research project challenges the official narrative about how people use artificial intelligence, showing patterns major AI companies have not disclosed.
Findings from the AI Observatory
The project, launched by independent researchers, analyzed user behavior across three leading AI models: Anthropic’s Claude, Google’s Gemini, and OpenAI’s ChatGPT. Company reports tend to emphasize professional or productivity-focused use cases, but the analysis revealed far more personal and sensitive interactions.
People were more likely to turn to Anthropic for coding, Gemini for social and roleplay uses, and ChatGPT for homework assistance. These results suggest that official reports may not capture the full range of usage.
The researchers note that without independent verification, the public cannot confirm whether the data released by AI companies is complete or representative.
Flock Safety’s surveillance decisions
Flock Safety, which operates a network of 120,000 automatic license plate readers across the U.S., recently announced platform updates to prevent misuse, including stalking. The changes arrive amid ongoing debate about balancing public safety and civil liberties.
Supporters claim that if the cameras help solve crimes like kidnappings, the trade-offs are justified. Critics argue that most license plate data goes unexamined, but this overlooks how Flock’s system was built. The company made deliberate choices about what data to collect, who can access it, how long it’s stored, and how widely it’s shared. Those choices set the boundaries between security and privacy.
Related: NASA Telescope May Spot Deadly Asteroids
A narrower system with stricter limits on data retention or access could achieve similar public safety goals while reducing risks. Flock’s updates address some misuse cases but do not alter the core architecture of its surveillance network. The debate over its role in policing will likely persist as long as the system’s design remains unchanged.
Transparency gaps in AI usage data
The differences between official reports and independent findings create concerns about transparency in the AI industry. When companies control the narrative around their tools, policymakers and the public may lack the full picture needed to regulate or understand the technology’s impact.
If many AI interactions involve personal or sensitive topics rather than workplace productivity, discussions about data privacy, content moderation, and mental health could shift. The analysis shows that available data is incomplete, and without independent oversight, this may not improve.
The findings also reveal model differences not always visible in company statistics. While ChatGPT dominates headlines for its versatility, users rely on it for specific tasks like homework, turning to other models for different needs. This segmentation could shape how AI tools develop, with each company adapting its product to observed behaviors—or those it chooses to acknowledge.
As AI becomes more embedded in daily life, unbiased usage data will become increasingly important. The project helps fill that gap, though much about the technology’s real-world applications remains unknown.
