Researchers Urge Independent AI Usage Data
Independent experts argue that current AI safety checks fail to capture real-world societal risks because they lack data on how millions actually use the tools.
Key points
- Independent researchers say auditing individual AI prompts fails to capture societal risks that arise from mass usage patterns.
- By mid-2025, 10 percent of global adults used ChatGPT, and 34 percent in six surveyed countries used AI weekly.
- Current usage dashboards are controlled by AI companies, preventing external experts from independently verifying safety trends.
Independent researchers argue that current AI safety checks fail to capture real-world societal risks. They propose creating a public tracking system similar to Google Trends. This would allow outsiders to see how people actually use chatbots.
The core problem is that testing individual prompts does not reveal large-scale patterns. A dangerous answer given to one person is a minor issue. The same answer given to millions creates a systemic risk. We need data on volume and frequency to assess true danger.
Current auditing methods miss the big picture
Many audits focus on specific questions asked to AI models. While this reveals biases or errors, it ignores who is asking. A harmful response is less concerning if it is rare. It becomes critical if millions of people encounter it.
Tech Policy Press notes that automating these tests often yields different results than real human usage. Companies may block unofficial automation. This means researchers cannot accurately measure how the average person interacts with the technology.
Rapid growth increases the need for transparency
Adoption of AI tools is expanding quickly. By July 2025, about 10 percent of adults globally used ChatGPT. In six countries, 34 percent of users check generative AI weekly. In the US, roughly half of adults have used chatbots.
This growth outpaces our ability to monitor it safely. People are increasingly exposed to AI summaries without actively seeking them. Understanding this shift requires detailed, up-to-date data on usage trends.
Company data lacks independent oversight
Major AI firms publish their own usage dashboards. OpenAI and Anthropic share internal metrics, but they control the narrative. These reports often focus on economic impacts rather than social safety. External researchers lack access to the raw data.
Even when partners like Harvard collaborate with companies, independence is limited. Experts warn that companies choose what to include and omit. A public, neutral tracking tool would solve this structural conflict of interest.
