Threat intelligence that reads text the way an analyst does
Percipion™ is a sovereign cognitive platform for sensitive language. It uncovers weak signals, hidden intent and emerging threats in text, offline, explainable, and under your control, built for law enforcement, defence, forensics and critical organisations that cannot send data to a third-party model.
Percipion
White Paper
Email us
Selected by PwC as one of 20 European companies for the Scale Defence & Resilience Tech programme at the Munich Security Conference, February 2026.
Most AI cannot be put in front of classified text
Keyword tools miss intent. Statistical models hallucinate, can be poisoned, and usually depend on foreign stacks. For investigators and operators, that is not a research inconvenience. It is an operational and legal risk.
What they need is different
- They need meaning, not only matches
- They need an audit trail, not a black box
- They need a system that runs on their infrastructure, on their data, with no outbound dependency
Percipion turns sensitive text into intelligence you can act on
Percipion is Aeteos’s symbolic cognitive platform. It does not imitate language with probabilities. It represents words, concepts, grammar, valence and context in knowledge schemas, then reasons over them.
It’s used to
- Read large volumes of messages, reports, open-source text and seized data
- Detect weak semantic signals before they become incidents
- Expose psychological pressure, grooming, coercion, hate and disinformation
- Produce explainable inferences for investigators, analysts and decision-makers
It can support a human decision, or trigger a reasoned response. It does not replace the analyst.
Two deployments, one engine
Percipion Sentinel
A light semantic filter and alert layer for live language flows.
Percipion Platform
The full cognitive engine: analysis, inference, weak-signal detection, reasoning and explainable reporting, on one server or across the organisation
From raw text to a reasoned assessment
Ingest
Internal communications, Email archives, open sources, messaging, seized devices, or case files. On your premises.
Represent
Percipion maps language onto knowledge schemas, structured symbolic memory of concepts, relations, emotions and threat patterns.
Infer
It reads nuance: algospeak, emoji, grammar, valence, psychological drift, and how those signals evolve.
Explain & Act
You get a traceable assessment: what was found, why it matters, and what to watch next. No fabricated sources. No hidden model weights.
Designed for sovereignty, not for the public cloud
Symbolic, not statistical
Knowledge-based reasoning. Deterministic enough to review. No hallucinations from next-token prediction.
Yours to operate
On-premise or fully offline. No data capture by Aeteos. No third-party components in the engine. No training on your corpora.
Hard to subvert
The system is not fine-tuned on incoming case data, so it cannot be poisoned the way a learning model can.
Explainable by construction
Inferences can be inspected. That matters for forensics, command, and the EU AI Act.
Built to stay
Developed in France, in portable C, for standard Linux environments. Predictable cost. Low energy use. No vendor model API behind the curtain.
Legal posture
Aligned with the EU AI Act and designed for NIS 2-era operators who must stay in control of high-risk processing.
Built for the people who protect others
Cognitive defence
Spot weak semantic signals, manipulation tactics, conspiracy narratives and coordinated disinformation while they are still forming.
Proactive threat detection
Watch language for grooming, radicalisation, hate, scams, coercion and hybrid-threat planning. Raise an early, reasoned alert.
Forensic analysis
Process seized messages and documents into explainable assessments. One laboratory use case: 373 pages reviewed in about a minute, against days of manual reading.
Insider & value-chain risks
Read internal and partner communications for pressure, disloyalty, exfiltration language and psychological drift, without sending the corpus outside the organisation.
Already in the hands of people who cannot name their cases
We publish outcomes without names, because that is how this work has to be done.
A national law-enforcement unit chose Percipion for a European effort against online hate speech, to process text offline and surface weak signals and valence.
A major European defence organisation used Percipion to analyse collected intelligence on influence activity and emerging threats.
A national forensic laboratory deployed Percipion offline against messages from seized devices.
A global NGO protecting children uses it to detect grooming behaviour in online interaction.
Recognition we will put our name on
Selected for PwC’s Scale Defence & Resilience Tech cohort (2026). Kick-off pitch at the PwC and Strategy& lounge during the Munich Security Conference, 13 February 2026.
Humans first. Then machines
Aeteos is a French family-owned company founded in 2016 by cognitive psychologists Mark and Béatrice Pohlmann. The work sits on a longer practice: understanding how people represent meaning, then building software that can do a careful version of the same thing.
Percipion comes from Latin percipio: to perceive fully, to grasp. Aeteos itself is Aether and Eos, clarity at first light. The point is not poetry. It is a design rule. The machine should help the protector see, not speak over them.
Request a briefing
If you want a demonstration, an evaluation on your own infrastructure, or a conversation about a specific mission set, write to us.
We do not put a public form on this page. One mailbox, handled by the team.