Explainability
Inside the model — representations, limitations and observable grounds for a decision.
- Transparency and explainability
POSITION
Independent AI observatory
Audit
the invisible.
AI systems are advancing faster than our collective ability to understand them. CSAEAI opens a field of research and action on their explainability, governance and audit.
CSAEAI / 2026Active research
SCROLL ↓
FINDING
Trust is not declared.
It is demonstrated.
Since 2024, CSAEAI has been building a rigorous approach to auditing artificial intelligence systems. We publish our doctrine, references and analyses — not our internal mechanisms.
Audit
Evidence, behaviour, risk and real-world consequences.
- Human rights and democratic values, including fairness and privacy
- Robustness, security and safety
Governance
Law, human accountability and decisions that can be explained and challenged.
- Inclusive growth, sustainable development and well-being
- Accountability
OECD principles · 2024OECD.AI ↗
REFERENCES
Transparent about the framework. Discreet about the method.
Explore six references here; the full atlas covers thirteen, from applicable law to strategies and voluntary frameworks. Citing them implies neither certification, endorsement nor automatic availability of an audit.
Explore all thirteen references →JOURNAL
Signals of the present.
Auditing AI: accountability, self-regulation and technological power
How can a technology be assessed when its producers also organise its evaluation? Gramsci, Schumpeter and Foucault illuminate the conditions of independent evidence and responsible auditing.
ANALYSIS · 7 September 2026Auditing AI agents: what the Hugging Face incident reveals
The Hugging Face incident makes the case for auditing agents’ actions, not just their final answers. CSAEAI examines transparency, reliable records and the conditions for independent scrutiny.
The future is already running.