Blog

AI can generate a convincing Amsterdam street that never existed, then fail to recognize its own fiction when asked to geolocate it. This experiment shows why visual plausibility is no longer proof. For OSINT practitioners, the lesson is clear: seeing is not believing. Only provenance, source tracing, corroboration, and disciplined geolocation tradecraft can separate synthetic evidence, from reality in an internet where images can lie with confidence and false places can feel

Evidence poisoning is the contamination of the open web with synthetic, manipulated or amplified material that later appears useful as investigative evidence. In an age of AI slop, fake personas, recycled footage and algorithmic summaries, OSINT practitioners must ask not only whether a source can be found, but whether it was manufactured to be found.

AI has made it easy for OSINT practitioners to build and install investigative tools. It has not made it safe to trust them. This blog examines the real risk behind vibe coding, viral GitHub apps, and useful-looking platforms that may expose workflows, distort analysis, or reveal investigative intent. In OSINT, the danger is not only bad code. It is trusting software you do not fully understand.

This is not just random extremism. It is a pipeline. From open platforms to closed chats, vulnerable people can be spotted, shaped, tested, and pushed toward real-world violence as disposable proxies. This piece examines how that works, why Jewish sites and dissidents are attractive targets, and how OSINT investigators can spot the runway before the fire starts.