Cambridge Analytica 2.0 – When the AI Assistant Becomes a Intelligence Graph

The piece warns that modern AI assistants, when given legitimate, broad access across email, finance, engineering, HR, supplier systems, calendars and documents, can join harmless fragments into a coherent strategic map of a company - replicating the Cambridge Analytica pattern of value coming from connections rather than raw items. An attacker who compromises such an assistant can reconstruct future products, launch dates, pricing, supplier dependencies, weaknesses, acquisition plans and more with high accuracy, and then export or act on that reconstruction. The workflow is explicit: permitted access gathers fragments, AI inference creates a new strategic picture, that inference becomes persistent memory or is shared, and then tool-enabled actions turn inference into real-world consequences. Existing logs and access tokens can read “everything normal” even while the combination and resulting actions produce cascading business collapse - lost market share, supplier disruption, IP exposure, regulatory and reputational damage.

The prescription is technical, not just policy: organisations must prevent joining where not authorised, separate inference from authority to act, enforce containment, segmentation and non-joinability, control AI memory and output release, and require independent checks at the action boundary. The central assurance question for any board or regulator is concrete: if the AI is compromised, how much of the enterprise can an attacker reconstruct, export and operationalise before an independent technical control halts it? Practical auditability, provenance of reads and inferences, per-action verification, and protection against compromised connectors are mandatory controls listed for CISOs and regulators.