A working definition
Open-source intelligence — OSINT — is the practice of collecting, analysing, and verifying information that is publicly available, then producing findings reliable enough to act on. “Open source” here has nothing to do with software; it means the source is open: news reporting, public records, official statements, corporate disclosures, social media, academic work, public archives, satellite imagery.
The discipline is old, but it has changed character. The constraint used to be access — finding the information at all. Today the constraint is volume and verification: the information is abundant, much of it is unreliable, and the hard work is establishing what is actually true.
Where it gets applied
The same discipline shows up across very different fields. The questions differ; the method is the same — track the source, corroborate the claim, keep the audit trail. A few concrete examples of where it is now routine:
- · Environmental monitoring — tracking deforestation and water-quality change from open satellite, NGO, and news sources; comparing corporate environmental commitments against what disclosures actually report
- · Urban planning — spotting large developments before they are formally announced via planning-board minutes and press, and comparing planning decisions across cities
- · Due diligence and KYC — building entity-and-ownership pictures from corporate registries, regulatory filings, and adverse media for a counterparty or asset
- · Public administration — monitoring procurement, budget execution, and council votes across jurisdictions over time
- · Infrastructure analysis — following large-project announcements and financing flows across regions and consolidating contractor information from procurement portals
- · Journalism and academic research — field-mapping a topic across thousands of public sources, with every claim carrying its provenance
Where the information comes from
OSINT draws on whatever the public record contains. In practice that usually means a mix of:
- · News and wire reporting, often across many languages
- · Official sources — government releases, court filings, regulatory and corporate disclosures
- · Public statements from organisations, institutions, and industry bodies
- · Public social media — posts, images, and video from identifiable accounts
- · Public archives — encyclopedic references, archived web pages, historical records, satellite imagery
How OSINT differs from classified work
Classified work relies on sources the public cannot see and, often, cannot check. OSINT's defining property is the opposite: because the sources are public, the findings can in principle be independently verified by anyone. That is a strength — it makes OSINT findings usable in research, journalism, regulatory work, and any setting where the analysis has to stand up to a second reader — but only if the provenance is preserved. An OSINT finding without its sources attached is just an assertion.
The verification problem
The central difficulty in OSINT is not collection; it is corroboration. A single source reporting something is a claim, not a fact. Establishing reliability means finding independent sources that agree, checking that they are genuinely independent rather than echoing each other, weighing source reliability, and keeping a trail of how the conclusion was reached.
Standards exist for doing this rigorously. The Berkeley Protocol on Digital Open Source Investigations is the most developed of them — it writes down a consistent methodology for collecting, preserving, and verifying open-source material, and the principles travel across sectors.
Why tooling matters
At small scale, OSINT can be done by hand. At the scale the modern public record demands, it cannot. The work of reading thousands of sources, extracting claims against a consistent schema, deduplicating, and corroborating is mechanical — and doing it by hand is where coverage gaps and provenance failures come from.
Generative AI is what changed the economics here, but AI does not replace OSINT methodology — it makes scaling possible while the methodology preserves the discipline. The mechanical reading is automated: AI extracts typed claims out of unstructured prose, line by line, across volumes a human team could never touch. The methodology stays around the AI: the original source URL, capture timestamp, content hash, the raw extraction payload, and corroboration across independent sources are recorded by construction at the moment of extraction, so the provenance trail is built in rather than reconstructed later. The investigator's judgement — what to trust, what to corroborate, what to call a finding — stays human. What the AI removes is the unreadable middle layer that used to swallow whole research teams.
A structured-intelligence workspace exists to do the mechanical layer reliably, so the analyst's time goes to the judgement only a human can exercise: source assessment, corroboration decisions, and the operational decision the work was for.