AI-P04 — AI for Intelligence Analysis and OSINT

Course Code: AI-P04  |  School: School of Artificial Intelligence  |  Cluster: Level 3 — Applied & Sectoral Practice

Level: Intermediate  |  Duration: 5 weeks · 18–22 learning hours  |  Language: English  |  Certificate: Professional Certificate (non-degree)  |  Format: Self-paced with AI support under human supervision

Overview

Open-source intelligence work has always been constrained by the volume of material a human can read and the difficulty of establishing that a source is what it claims to be. Machine assistance changes the first constraint substantially and the second not at all — while simultaneously making it far cheaper for an adversary to manufacture convincing material.

This course teaches the analyst’s response to that combination. Collection and triage at scale, translation and entity extraction, and pattern surfacing are treated as legitimate and useful applications. Verification, attribution and assessment are treated as irreducibly human, and the standard of proof for each is raised rather than lowered because synthetic material is now abundant.

The course is explicit about limits. It teaches verification of publicly available material for legitimate research, journalism and institutional risk work. It does not teach circumvention of access controls, deanonymisation of private individuals, facial recognition against persons, or any technique whose purpose is covert surveillance. Learners work only with lawfully accessible sources and are required to document the ethical basis of each collection decision.

Learning outcomes

On completion, a successful learner will be able to:

1. Plan a collection effort against a defined question, with lawful, documented sourcing and an explicit ethical basis.
2. Use machine assistance for triage, translation and entity extraction while recording provenance for every item retained.
3. Verify content, source and context using established techniques, and state the residual uncertainty.
4. Assess the likelihood that material is synthetic or manipulated, and act appropriately when the answer is uncertain.
5. Distinguish correlation surfaced by a tool from an attribution claim that evidence can support.
6. Write an assessment with calibrated confidence, a source appendix, and an explicit statement of what was not established.

Who this course is for

Analysts in institutional risk, compliance and security functions; investigative researchers and journalists; policy and strategic analysis units; and academic researchers working with open-source material.

Prerequisites

Prior study of OSINT method or equivalent professional experience. AI-F06 required in substance: the verification discipline is assumed. Learners must agree to the course ethics undertaking before commencing the assessed work.

Syllabus

Module 1 — Question, legality and ethical basis

Focus. Framing a question that open sources can answer, establishing the lawful and ethical basis for collection, and defining boundaries before any collection begins.

Lessons. 1.1 What open sources can and cannot answer. 1.2 Lawful access and terms of service. 1.3 Proportionality and the individual’s privacy interest. 1.4 Boundaries recorded in advance.

Core reading. Michael S. Lowenthal, Intelligence: From Secrets to Policy, 9th edition (Thousand Oaks: SAGE/CQ Press, 2022), chapters on collection disciplines and open sources. Peter Gill & Mark Phythian, Intelligence in an Insecure World, 3rd edition (Cambridge: Polity Press, 2018), chapters 6–7.

Deliverable. Collection plan with recorded legal and ethical basis and stated boundaries.

Module 2 — Triage, translation and extraction at scale

Focus. Where machine assistance genuinely helps: reducing a large corpus to a reviewable set, working across languages, extracting entities and relations for later human checking. Provenance capture as a non-negotiable step.

Lessons. 2.1 Triage without discarding the important. 2.2 Machine translation: utility and evidentiary limits. 2.3 Entity and relation extraction. 2.4 Provenance, hashing and archival capture.

Core reading. Craig Silverman (ed.), Verification Handbook (Maastricht: European Journalism Centre, 2014), chapters 1–3.

Deliverable. Triage record for a real corpus, with provenance captured for every retained item.

Module 3 — Verification of content, source and context

Focus. The three separate questions: is the artefact authentic, is the source who they claim to be, and is the context as presented. Geolocation, chronolocation and cross-source corroboration.

Lessons. 3.1 Content authenticity. 3.2 Source identity and history. 3.3 Context and the recycled-image problem. 3.4 Corroboration and what counts as independent.

Core reading. Silverman (ed.), Verification Handbook, chapters 4–7. Eliot Higgins, We Are Bellingcat: An Intelligence Agency for the People (London: Bloomsbury, 2021), chapters 3–5.

Deliverable. Verification dossier for three items, each with a stated residual uncertainty.

Module 4 — Synthetic and manipulated material

Focus. What current generation and manipulation techniques can produce, why detection is unreliable as a sole basis for a judgement, and how to reason under the assumption that any single artefact may be fabricated.

Lessons. 4.1 Capabilities of current generation techniques. 4.2 Forensic indicators and their limits. 4.3 Detection tool unreliability. 4.4 Reasoning when authenticity cannot be settled.

Core reading. Luisa Verdoliva, “Media Forensics and DeepFakes: An Overview”, IEEE Journal of Selected Topics in Signal Processing 14, no. 5 (2020): 910–932. Laura Weidinger et al., “Taxonomy of Risks Posed by Language Models”, Proceedings of FAccT (ACM, 2022).

Deliverable. Authenticity assessment of a supplied item set, including at least one case reported as unresolved.

Module 5 — From pattern to assessment

Focus. The gap between a correlation a tool surfaced and an attribution the evidence supports. Writing the assessment with calibrated confidence and an explicit account of what was not established.

Lessons. 5.1 Correlation, coincidence and attribution. 5.2 Alternative explanations. 5.3 Calibrated confidence language. 5.4 Stating what you did not establish.

Core reading. Randolph H. Pherson & Richards J. Heuer Jr., Structured Analytic Techniques for Intelligence Analysis, 3rd edition (Thousand Oaks: SAGE/CQ Press, 2020), chapter 7. Richards J. Heuer Jr., Psychology of Intelligence Analysis (CIA CSI, 1999), chapter 8.

Deliverable. Final submission: assessment with source appendix, calibrated confidence and a negative-findings section.

Assessment

Pass mark 70 per cent. All assessed components must be attempted. Every mark in this course is issued by a human assessor; no assessment outcome is generated automatically.

Rubric criteria

Each assessed artefact is marked against four criteria at four levels (distinction, pass with merit, pass, fail).

1. Ethical discipline: is every collection decision within the stated lawful and proportionate boundary, and documented?
2. Verification rigour: are content, source and context treated as separate questions with separate evidence?
3. Epistemic honesty: is uncertainty reported rather than resolved by assumption, including where the answer is unresolved?
4. Attribution restraint: does the assessment stop where the evidence stops?

Reading list

Core. Craig Silverman (ed.), Verification Handbook (Maastricht: European Journalism Centre, 2014). Randolph H. Pherson & Richards J. Heuer Jr., Structured Analytic Techniques for Intelligence Analysis, 3rd edition (Thousand Oaks: SAGE/CQ Press, 2020).

Discipline and context. Michael S. Lowenthal, Intelligence: From Secrets to Policy, 9th edition (Thousand Oaks: SAGE/CQ Press, 2022). Peter Gill & Mark Phythian, Intelligence in an Insecure World, 3rd edition (Cambridge: Polity Press, 2018). Eliot Higgins, We Are Bellingcat (London: Bloomsbury, 2021).

Peer-reviewed. Luisa Verdoliva, “Media Forensics and DeepFakes: An Overview”, IEEE Journal of Selected Topics in Signal Processing 14, no. 5 (2020). Laura Weidinger et al., “Taxonomy of Risks Posed by Language Models”, Proceedings of FAccT (ACM, 2022).

All items are published works identifiable by author, title and publisher. Learners obtain them through an institutional library or the publisher. The Academy does not distribute copyrighted texts.

Academic integrity and use of AI

Generative tools may be used in producing assessed work under three conditions. Use must be disclosed in a short statement appended to each submission, naming the tool and the task it performed. Any factual or technical claim originating from a generative tool must be verified against a citable source before it enters assessed work, and the verification must be evidenced. The analytical judgement in each artefact must be the learner’s own and must be defensible in a short follow-up. This course does not teach circumvention of access controls, deanonymisation of private individuals, facial recognition against persons, or covert surveillance. Assessed work involving any of these is refused and the learner is withdrawn.

Instructor: pending owner confirmation. Pricing: pending owner approval. Reference list verified against publisher records; any later addition is marked for verification before publication.