ChatGPT users are figuring out the location of photos using o3's image-analyzing capabilities paired with its web search functionality, raising privacy concerns
There's a somewhat concerning new trend going viral: people are using ChatGPT to figure out the location shown in pictures.
Context & Ripple Effects
This is an early example of a general-purpose AI workflow creating a privacy risk through capability combination rather than a single new feature. A later hands-on account showed o3 using code to inspect license plates while making a location guess, underscoring how image reasoning can draw on multiple clues in a geolocation workflow that examined license plates.
The issue sits alongside OpenAI's separate experiments with watermarks for images made through ChatGPT, which address image provenance but not the exposure of location clues in ordinary photos. It also precedes OpenAI's removal of a ChatGPT experiment that made conversations searchable, another instance in which product distribution created an unexpected privacy surface.
First-order effects
- People who share photographs can face a new exposure path: other users may combine visual details and web results to infer where an image was taken.
- OpenAI's image-analysis and search combination becomes subject to privacy scrutiny based on how users apply the workflow, not only on the accuracy of either component alone.
Second-order effects
- Other AI products that pair visual reasoning with live search will face the same question: whether their combined tools make location inference too easy for ordinary users.
- Image provenance measures such as ChatGPT's reported watermark testing may become more salient, but they do not directly resolve location inference from real-world images.
Third-order effects
- If such workflows become routine, multimodal AI safety assessments will need to evaluate end-to-end user tasks—image interpretation, code use, and search—rather than treating each capability as an isolated feature.
- The likely structural pressure is toward clearer privacy boundaries around consumer-grade investigative capabilities, especially as access to stronger image models broadens; the corpus does not establish which safeguards will prevail.
The trend: This is one data point in the shift from standalone generative AI features toward multimodal agents whose combined tools can turn public visual traces into actionable personal information.