Velqory Blog | AI, cybersecurity and regulation

Showing

Artificial intelligence

1–12 of 54

Current knowledge, organised by topic, date and practical consequence.

Conceptual scene of a technician examining wheat samples with a generic sensor at a quality-control bench. — Quality control as a visual example of a bounded applied-AI task. Generated scene; not an actual finalist or pilot.AI-generated image under Velqory editorial direction.

Lead analysis

Artificial intelligence05 Oct 20264 min read

Applied AI in Europe: buy a pilot, not a promise

European finalists offer a starting point for finding solutions. An SME needs a different dossier to decide: its own tasks, permitted data and outcomes someone can accept.

Read analysis

Person puts an authentication key in a small open safe beside a notebook without text. Generated conceptual scene; not documentation of an actual event, product or test.AI-generated image under Velqory editorial direction.

Artificial intelligence05 Oct 20264 min read

04/10 edition: the system prompt is not a secret vault

Making an instruction harder to extract does not fix a credential placed in context. Remediation starts by removing the secret and proving control outside the model.

Read analysis

Person compares two labels with printed shapes and bars beside a parcel and a printer. Generated conceptual scene; not documentation of an actual event, product or test.AI-generated image under Velqory editorial direction.

Artificial intelligence05 Oct 20264 min read

29/09 edition: validate the answer before turning it into action

An AI answer may look like text but end up interpreted as HTML, a query or a file path. Validation belongs at the destination, not in trust in the model.

Read analysis

Person retrieves an archive folder beside a locked cabinet and a scanner. Generated conceptual scene; not documentation of an actual event, product or test.AI-generated image under Velqory editorial direction.

Artificial intelligence05 Oct 20264 min read

24/09 edition: revoke document access in RAG as well

A restricted file should not reappear in an answer through its excerpts. Ingestion, retrieval and deletion need separate evidence.

Read analysis

Two people compare parcel photographs sorted into trays, with one photograph between them. Generated conceptual scene; not documentation of an actual event, product or test.AI-generated image under Velqory editorial direction.

Artificial intelligence05 Oct 20264 min read

19/09 edition: an acceptance corpus for classifying requests

Before routing requests automatically, an SME needs examples defining tolerable errors, ambiguity and who can correct the label.

Read analysis

Person places a sheet in a scanner, with documents separated in a tray and behind a transparent divider. Generated conceptual scene; not documentation of an actual event, product or test.AI-generated image under Velqory editorial direction.

Artificial intelligence05 Oct 20264 min read

14/09 edition: in RAG, a document does not give orders

A proposal assistant can consult a dossier without receiving authority from it to send files. The useful boundary separates retrieved content from authorised action.

Read analysis

Conceptual editorial illustration about artificial intelligence. — Conceptual image generated for this analysis; it does not document a real event.AI-generated conceptual image by Velqory.

Artificial intelligence12 Sept 20264 min read

AI Omnibus: review timelines and governance

The European update changes stated milestones for high-risk AI systems and warrants a prudent operational review.

Read analysis

Artificial intelligence11 Sept 20264 min read

EU AI: preparing evidence for oversight

Preparation for AI oversight should start with a map of responsibilities, evidence, and versions—not generic compliance assertions.

Read analysis

Artificial intelligence10 Sept 20264 min read

AI transparency before enforcement

Transparency should enter product, content, and service workflows as a European operational control.

Read analysis

Artificial intelligence09 Sept 20264 min read

AI transparency controls for the EU

Transparency for interactive systems and AI-generated content needs operational evidence, not merely product messaging.

Read analysis

Conceptual editorial illustration for “Generative AI in Public Administration: From Demonstration to Compliance Control”. — Conceptual image generated for this analysis; it does not document a real event.AI-generated conceptual image by Velqory.

Artificial intelligence08 Sept 20264 min read

Public AI pilots with verifiable controls

Turning generative-AI pilots into operational decisions requires evidence, clear scope, and compliance gates.

Read analysis

Conceptual editorial illustration for “AI: From Transparency Duty to Operational Proof”. — Conceptual image generated for this analysis; it does not document a real event.AI-generated conceptual image by Velqory.

Artificial intelligence07 Sept 20264 min read

AI: From Transparency to Operational Proof

The European Commission communication points to transparency obligations that need operational evidence, not only statements of intent.

Read analysis

How we publish

Each analysis links to the official source and says what changes in practice.

01Sources in view
Every analysis links the official documentation it relies on.
02Operational context
We translate change into decisions, owners and evidence.
03Continuous review
We update the analysis when facts or guidance change.
04Identified imagery
We disclose generated photography and credit documentary photography.

From context to execution

Is there an obligation or a system that needs to become operational?

We map the problem, write the scope and build the solution with acceptance criteria.

Talk to Velqory Explore services