AI-Powered OSINT for Journalists: Ethics and the Berkeley Protocol
AI helps journalists and researchers handle the volume of open source material: translating posts, transcribing video, clustering similar content and flagging possible duplicates. It does not replace verification.
Topic: AI in Cybersecurity in Cybersecurity & Ethical Hacking · Explainer
Also covers: Security Tools, Authorized Labs & CTFs
Quick answer: AI helps journalists and researchers handle the volume of open source material: translating posts, transcribing video, clustering similar content and flagging possible duplicates. It does not replace verification. The Berkeley Protocol, published by the UN Human Rights Office and UC Berkeley, sets minimum standards for identifying, collecting, preserving, verifying and analysing digital open source information, and those standards apply whether or not AI is used. Use AI to find and sort, then verify by hand, preserve originals with hashes and timestamps, record every AI step, minimise personal data, and protect the safety and wellbeing of sources, subjects and your own team.
Key takeaways
- AI speeds up discovery and sorting, not verification.
- The Berkeley Protocol gives a standard workflow for open source investigations.
- Preserve originals before analysing them with any tool.
- Avoid harm: identity mistakes and exposure can endanger people.
- Viewing distressing content needs psychosocial safeguards.
Table of Contents
- What Is the Berkeley Protocol?
- Where Does AI Fit in Each Investigation Phase?
- Why Is Verification Still Manual?
- How Do You Avoid Harm to People?
- How Do Investigators Protect Themselves?
- What Should You Record About AI Use?
- References
What Is the Berkeley Protocol?
The Berkeley Protocol on Digital Open Source Investigations was launched on 1 December 2020 by the Office of the UN High Commissioner for Human Rights and the Human Rights Center at UC Berkeley School of Law. Its UN publication record lists it as ST/HR/PUB/20/2. It was written for international criminal and human rights investigations, but its standards are widely used by newsrooms and academic researchers.
The protocol is tool neutral. It does not ban or endorse AI; it asks that every method be documented, repeatable and proportionate to the purpose. That makes it a practical yardstick for deciding whether an AI step is acceptable in your workflow.
Where Does AI Fit in Each Investigation Phase?
| Phase | Useful AI help | Human responsibility |
|---|---|---|
| Online inquiry | Translate search terms, suggest keywords | Define scope and purpose |
| Preliminary assessment | Cluster near-duplicate posts | Judge relevance |
| Collection | Transcribe audio, extract text | Capture originals with metadata |
| Preservation | Little role | Hash files, keep chain of custody |
| Verification | Suggest leads for location or date | Confirm with independent evidence |
| Analysis | Summarise large sets, draft timelines | Draw and own the conclusions |
Why Is Verification Still Manual?
Generative models can produce confident but false statements, which NIST AI 600-1 calls confabulation. They also cannot tell you who first posted a video or whether it was recycled from an older event. Verification means finding the earliest upload, checking the uploader, matching landmarks and weather, and looking for C2PA content credentials where present. AI detectors for synthetic media help as one signal only. Our guide to AI tools for OSINT and how to verify them covers tool testing.
How Do You Avoid Harm to People?
| Risk | Safeguard |
|---|---|
| Wrong identification | Require two independent identifiers; never name on AI output alone |
| Exposing victims or witnesses | Blur faces and remove location details before publishing |
| Over-collection of personal data | Collect only what the story needs, as the DPDP Act 2023 expects |
| Uploading sensitive material to cloud AI | Use approved or local tools for sensitive files |
For broader ethical questions, see ethical implications of AI-powered OSINT.
How Do Investigators Protect Themselves?
The Berkeley Protocol covers digital, physical and psychosocial safety. Use separate research accounts and devices, keep software updated, and avoid logging in to personal accounts during research. Graphic content causes real stress: limit exposure time, turn off autoplay, view footage without sound first, and make support available. If research uncovers fraud or threats, report through cybercrime.gov.in or the 1930 helpline in India.
What Should You Record About AI Use?
- Tool name, version and date of use.
- The input given and the output received.
- How each output was checked and by whom.
- Which conclusions rest on AI-assisted steps.
This record supports editorial review and fits the governance approach of the NIST AI RMF. WebAsha offers OSINT and digital investigation training.
References
Frequently Asked Questions
What's Your Reaction?
Like
0
Dislike
0
Love
0
Funny
0
Wow
0
Sad
0
Angry
0