Source
European External Action Service (2025), LIT-084, 3rd EEAS Report on Foreign Information Manipulation and Interference Threats: Exposing the architecture of FIMI operations (March 2025), read in NOTE-060. Page numbers are the printed ones.
What was actually shown
The sample (pp. 6, 9–11). 505 incidents detected and analysed by EEAS strategic monitoring in 2024: about 38,000 unique channels on 25 platforms, more than 68,000 observables, 90 countries and 322 organisations targeted. Disposable bot and coordinated-inauthentic accounts were 28,000 of the 38,000 channels (73%). The report says the incidents reflect “known outlets related to overt FIMI or independently attributed operations by selected actors and on priority issues”, do not cover all regions or languages, and “should not be used to draw conclusions about general trends”.
The graph (p. 20). Nodes are channels; an edge is shared involvement in an incident. Only channels in more than one incident with at least two connections are kept: 2,055 nodes and 8,056 edges. By attribution: Russia 20%, China 3.5%, non-attributed 76.5%.
The Exposure Matrix (pp. 14–19). Four categories by the strength and kind of evidence tying a channel to a state: official state channels; state-controlled outlets (RT, Sputnik, CGTN and mirrors sharing their infrastructure); state-linked channels, covert but attributed on high-confidence, mostly proprietary or classified indicators; and state-aligned channels, not attributed but showing “strong, repetitive patterns of alignment” on medium-confidence, mostly open indicators. Narrative alignment alone is not enough, over-attribution is to be avoided (p. 14), and classification is “not static”. The final public attribution is “often political” (p. 19).
What could have come out otherwise. Attributed channels could have been the bulk of the recurring infrastructure. In this sample they were not.
What this does not say
- That three-quarters of FIMI channels are unattributed. The 76.5% is a share of the filtered graph, excluding the 28,000 inauthentic accounts and every single-incident channel. The conclusions (p. 38) restate it as “a sizeable majority (76.5%) of the channels investigated”, which the body does not support (NOTE-060, corrections).
- Anything about prevalence or trend. Every count is a count of what the EEAS chose to monitor.
- That the Matrix sorts channels reliably. No inter-rater test and no worked example of weighing indicators for a channel. Category 3 rests partly on evidence not shown (C1, moderate as a proposal).
- That China’s network is centralised and Russia’s decentralised. Read off the picture, with no metric (C6, weak–moderate).
- Anything about impact. The report says “true reach and impact” remain unclear (p. 11). About 41 incidents involved AI, and AI’s use “does not necessarily translate into a greater impact”.
- That its figures reconcile. The per-platform channel counts add to about 63,000, against 38,000; X’s 88% has no stated base; the foreword and body differ on countries and organisations targeted.
Connections
- QUESTION-070 asks whether an organized whole manipulates when no member intends to, and names this report’s tiered network as one of its two cases. This account says what the network looks like in the EEAS’s sample: most of its recurring channels are graded only as state-aligned. It does not answer the question, which turns on what manipulation is.
- THEORY-651 (DISINFOX, LIT-083) carries incidents in STIX with attribution as one ungraded edge; the Matrix is the graded layer that schema lacks.
- THEORY-488 and THEORY-551 give the Russian side. The report names “information confrontation” as central to Russian doctrine (p. 12) without citing a Russian text; Fridman (LIT-088) cites Russian work on it. The foreword’s FIMI used “to lay the way for kinetic action” (p. 2) parallels Clark’s subordination of kinetic action to information.