Weak signals are not invisible. They are ignored.
What Guido Palazzo, professor of business ethics at the University of Lausanne (UNIL), documents in his work on organisational blindness — that progressive drift where competent people stop seeing what is plainly there — is exactly the mechanism at work with weak signals. Not a failure of values. A failure of perception, installed by context.
These signals exist and circulate in your HR data, your collaborative tools, your internal exchanges. But performance pressure, (over)confidence in algorithms, and the speed imposed by AI narrow the focus until they become invisible.
Everything rests on this perception. Decisions are made based on what we think we see. Hackers and manipulators know this. They don’t attack your systems — they attack your reading of reality. AI amplifies this fragility at a speed organisations have not yet absorbed.
Ignoring them feeds the crisis. Reading them means taking back control.
- State of play 2026: ize does not protect against blindness
- The four main families of weak signals, and why AI makes them explosive
- Integrating the weak signals in dashboards, provided that you change your posture and mindset
- A concrete example of integrating weak signals in dashboards
Size does not protect against blindness
Every organisation is affected… but not in the same way.
Large organisations have the tools, the processes, the budgets. And yet, in most post-crisis analyses — whether cyber, health-related or reputational — the signals existed. They did not circulate, or were neutralised by the context itself. Organisational complexity does not protect against blindness. It can even create it.
SMEs, on the other hand, often lack the time and resources to formalise vigilance. But they sometimes retain what large structures lose: the proximity that still allows them to listen.
What changes the equation: many SMEs are part of the direct ecosystem of large organisations — subcontractors, service providers, partners. These organisations impose growing requirements on them in terms of AI risk management, without always ensuring they have the real means to comply. Yet a vulnerability in a small actor is a gateway into the large organisation’s system. The security of the whole depends on the most exposed link — not the strongest.
In both cases — large organisation or SME — the finding is the same: risks are identified, but the collective capacity to face them remains largely insufficient. Not for lack of resources, but through a failure of perception. This is precisely where weak signals come into play.
→ To assess your organisation: cyberAI crisis readiness diagnostic
What weak signals are, and why AI makes them explosive
Weak signals are the early manifestations of a system under tension. They are the micro-signs of a progressive deterioration in the collective capacity to perceive, interpret and decide correctly. They always precede the visible crisis. ENISA, the WEF, and ANSSI document this convergently. Major crises do not erupt suddenly. They settle in silence, signal after signal ignored.
For Palazzo, this silence is not an accident. It is produced by the organisational context itself, which progressively erodes the capacity to see what is inconvenient.
AI does not change this mechanism. It accelerates, amplifies and obscures it. It fabricates false signals, drowns real ones, and compresses reaction time until it is almost zero. What was counted in days is now counted in minutes.
Karl Weick, professor of psychology at the Ross School of Business at the University of Michigan, puts it this way: “Sensemaking is about enlarging small cues. It is a search for contexts within which small details fit together and make sense.” This process — sensemaking — is precisely what organisations lose when they are under pressure, fragmented, or blinded by their own tools.
It is from this framework that we distinguish four dimensions of the weak signal: perception, the human factor, the algorithmic and AI factor, and the organisation itself. Not four watertight compartments, but four layers of the same system under tension — all traversed, to varying degrees, by AI.
1️⃣ Perception weak signals, or what the world says about you
Irony, rumours, visual misappropriations or memes, sarcasm, ambiguous comments, on social media or in the corridors. These signals do not directly attack the facts. They attack the credibility, legitimacy and trust placed in the organisation’s words.
These signals do not directly attack the facts. They attack credibility, legitimacy, the trust granted to the organisation’s voice. Distrust sets in before any formal incident.
With AI, this dynamic changes in nature. It is no longer just opinions circulating — these are fabricated proofs. ❕ A vishing attack impersonating an executive announcing a product withdrawal, ❕ a perfectly convincing fake interview, ❕ astroturfing campaigns — automated accounts simulating a grassroots movement that looks authentic. With AI, attackers no longer seek to amplify an existing rumour. They create one from scratch.
When a crisis erupts in this context, the ground has already been prepared. The official communication arrives too late, in an environment that has become hostile — and facing “evidence” that facts alone are no longer enough to refute.
2️⃣ Human weak signals, or what your teams are feeling
Chronic fatigue, self-censorship, unusual silences, gradual disengagement, increasing absenteeism, number of burnout cases, sense of injustice.. These signals are well documented in post-crisis analyses. The pattern is recurring: the alerts existed but did not circulate.
Palazzo’s analysis: people did not decide to stay silent. It is the organisational context that progressively made silence safer than speaking up.
AI adds a new and less visible layer: professional grief. The EU-OSHA OSH Pulse 2025 documents this: workers exposed to intensive use of algorithmic tools report more stress, fatigue and loss of meaning. Some develop a feeling of dispossession — their judgement short-circuited, their expertise rendered obsolete by an opaque system. They do not resign. They do not complain. They progressively stop transmitting their knowledge, taking initiatives, flagging what is going wrong. Silent compliance in place of engagement.
This is particularly critical in sectors where expert knowledge is a strategic asset — regulatory, pharmacovigilance, formulation. When these experts go silent, the organisation does not know until the knowledge is gone.
In many cases, the crisis worsens because teams have stopped talking — because leaders have only taken data into account, not behaviours or feelings.These signals are well known from studies on organisational accidents and systemic crises.
3️⃣ Algorithmic weak signals and AI, or what systems produce
Abnormal amplification of content, credible fakes, deepfakes, automated reputation attacks, biased or misinterpreted algorithmic recommendations. The convergent reports from ENISA and the World Economic Forum document it: by accelerating, amplifying and obscuring information flows.
I distorts the crisis curve itself. The peak becomes more brutal, the false trough creates an illusion of resolution where the crisis continues to unfold, and the reputational impact persists long after the organisation believes it has regained control. Reaction time is no longer counted in days but in minutes.
This is a subject we develop elsewhere.
One thing to remember: when the weak signals of this family are ignored, the window of action closes before anyone realises it was open.
4️⃣Organisational weak signals, or what your structure allows or prevents you from seeing
Recurring dysfunctions, information hoarding, persistent silos, wolf packs, deferred arbitrations, blurred responsibilities, ambiguous rules, ethical compromises. These signals reveal a structural inability to see, decide and act in time.
This is the terrain Palazzo documents most precisely in The Dark Pattern. Under the pressure of a short-term profitability ideology, organisations have progressively shifted responsibility downward. “You are responsible for your remit, manage it like your own business” — BUT without transferring the decision-making power that goes with it. Result: a chasm between stated responsibility and real power, filled by tacit arrangements, euphemisms that normalise drift, and organised silence.
Post-crisis analyses — cyber, industrial, healthcare, financial — show a constant pattern: the information existed. The organisational system did not allow it to circulate or be effectively taken into account. Result: decisions accumulated by default, partial or retrospective indicators, critical chains of events that no one has overall mastery of.
Agentic AI adds a new paradox: the more agents appear to function effectively, the less their governance is questioned. Entire processes are automated without anyone being formally responsible for their outputs. Decisions are made on the basis of algorithmic recommendations whose origin is opaque. It is ghost governance — invisible precisely because it resembles efficiency.
When the organisation is already slow, fragmented or blind, AI accelerates bad decisions, amplifies blind spots, and turns tacit arrangements into de facto policies.
Integrating weak signals into dashboards: changing mindset before changing tools
Weak signals already exist in HR data, collaborative tools, digital usage, internal exchanges, and IT processes. What is missing is the mindset to read them.
Traditional dashboards seek to confirm that everything is working. Weak signals indicate precisely where something is beginning to crack — long before the crisis becomes visible, nameable, or media-exploitable. Two incompatible postures, unless you consciously decide to make room for the second.
Taken in isolation, these signals seem trivial: declining participation, shorter responses, lengthening delays, fewer cross-functional initiatives. Taken together, over time, they reveal collective fatigue, a loss of trust, a progressive withdrawal of frontline leadership. This is not a question of complex tools. It is a question of decision: accepting that some indicators do not serve to optimise, but to alert.
Two blind spots persist in most current dashboards.
- Reading aggregate rates without disaggregating them. Stable absenteeism overall can mask an explosion of long-term absences in a specific population. The EU-OSHA OSH Pulse 2025 documents this: long-term absences of more than 90 days now account for more than half of total absenteeism. Reading the aggregate rate means missing what is really happening — the rise of psychosocial risks, silent exhaustion in specific populations. The aggregate figure reassures. It does not alert.
- The total absence of an AI reading layer. Tool adoption rates by function, gaps between declared and actual usage, signals of circumvention of official tools. These indicators are beginning to be recommended by players such as Workday or Deloitte, but remain absent from the vast majority of operational dashboards. Yet these are precisely the signals that reveal whether an organisation truly masters its AI transformation — or is experiencing its effects without seeing them.
Making weak signals visible, a concrete example
Making visible what we usually choose not to see.
The table below offers a new reading of indicators that already exist — often collected for other reasons, but rarely interpreted as warning signals. This is the change of posture in action.
Modern prevention is not about stacking KPIs, but about linking micro-variations across several layers of the system: human, social, informational, emotional.
| Domain | Signal to observe | Where to detect it | What it really means |
|---|---|---|---|
| Internal Climate | Drop in survey participation Shorter, more neutral responses | HR tools Anonymous forms | The “Care” is declining. The emotional fabric is thinning. |
| Digital Interactions | Rise in emails with +3 recipients Longer response times | Internal messaging | Trust is declining. Communication is becoming defensive |
| Collective Energy | Rise in micro-absenteeism Lower activity on Mondays and Fridays | HR Data Badge logs System log-ins | Energy is draining away. The collective is tiring. |
| Governance Transparency | Rise in anonymous submissions or “nameless questions” | Feedback tools Internal Meetings | People are still talking — but no longer openly. Rumours. |
| Cohesion & Belonging | Fewer reactions on internal posts Decline in cross-team initiatives | Intranet Yammer Teams | The “we” becomes “every man for himself”. |
| Proximity Leadership | Cancelled meetings Fewer informal exchanges | Agenda On-site observations | Management is losing touch with reality. |
| Perception of Security | Rise in “unconfirmed” incident tickets or false cyber alerts | Helpdesk / IT | The emotional climate of safety is deteriorating. |
| AI adoption & shadow IT | Gap between official AI tools and undeclared tools Rise in unauthorised access | IT logs / Usage audits | The organisation is bypassing safeguards. AI governance is lagging behind actual usage. |
👉 These signals complement classic indicators. They capture the living dimension that figures forget: trust, fear, fatigue, loyalty.
The HSE (Health and Safety Executive) identifies classic manifestations of weak signals around workplace ill-being: withdrawal, isolation, lateness, demotivation, loss of concentration, etc.
3 things to remember
1. Technology will not save you. It is necessary — and it is used by humans, for humans. They are the ones who make the difference between a managed crisis and a suffered one.
2. The short term costs more than the long term. Ignoring weak signals today means paying for the crisis tomorrow — with interest.
3. AI is an opportunity. It forces us to look at what we were avoiding: people, organisation, governance. It is not an additional constraint. It is an invitation to build more lucid organisations.
And you, where are you with weak signals?
To go further… contact us.
