Palantir and Intelligence Sharing in Counter-Terrorism

Every major terrorist attack in a Western country over the past two decades has been followed by an inquiry, and the inquiries say substantially the same thing. The information existed. It was held by more than one organisation. Nobody assembled it. The recommendation is almost always better information sharing, and the recommendation is almost always only partially implemented.

The reason is not institutional obstinacy, although that exists. It is that the barriers are legal, structural and in some cases entirely proper, and that a system which removed them all would be unacceptable in a democracy.

The barriers worth understanding

Police hold material gathered under criminal procedure, with rules about disclosure and about its use in court. Security services hold intelligence gathered under different authorities, often with source protection obligations that override operational convenience. Immigration, financial regulators and local authorities each hold material under their own statutory basis.

Combining them is not a question of connecting databases. It requires a legal basis for the combination, a purpose that justifies it, and a record that demonstrates both. A platform that makes the technical combination easy while leaving those questions open is a liability, not a capability.

Federated matching rather than pooling

The architecture that tends to survive legal scrutiny does not pool material centrally. Each organisation retains its holdings and the shared layer answers a narrower question: does another participant hold material relating to this subject, and if so who should be contacted.

That is a considerably weaker capability than full fusion and it addresses the specific failure the inquiries describe. An officer investigating a subject learns that another agency has an interest, and a human conversation follows under the appropriate authority. The system has prompted the connection without transferring the material.

It also fails more safely. A pooled repository is a single point of compromise and a single point of scope creep, and both of those risks compound over time in ways that are difficult to reverse once the repository exists.

Source protection constrains the design

Human source material carries obligations that are not negotiable, and in some cases revealing the existence of reporting is itself disclosive. This is why intelligence organisations are cautious about systems that make their holdings discoverable, and the caution is well founded rather than territorial.

Separating assessment from source material is what makes participation possible. A partner can receive a judgement, with stated confidence, without the reporting behind it. Keeping product, evidence and source as distinct linked objects, each with its own handling, is the structural requirement that enables this, and systems that store assessments as documents cannot do it.

Prediction is the wrong framing

Attempts to score individuals for likelihood of violence have a poor record and a worse ethical position. The base rate is extremely low, which means that even a highly accurate classifier generates overwhelmingly more false positives than true ones, and each false positive is a person subjected to attention they did not earn.

The defensible use of these systems is retrospective and investigative: given a subject who has come to attention through lawful means, assemble what is known quickly and accurately. That is a genuine improvement on the status quo and it does not require predicting anything about anyone.

Oversight is what keeps the capability

Append-only audit of who queried what, under which authority and for what stated purpose is not a constraint on these systems. It is the condition of their continued existence. Capabilities that cannot be inspected get withdrawn after the first scandal, and the withdrawal is usually indiscriminate.

AIC builds in this space with evidence, provenance and recorded release as design constraints. Our PAL.NET SDK gives .NET teams typed access to Foundry and their own ontology where that is the platform in use.