WebAI
WebAIVerify
Knowledge preview

Verify

Evidence, provenance and editorial accountability.

C2PAStandardsTrust
UNDERSTAND THE TOPIC

A citation is not automatically relevant evidence. Provenance metadata does not guarantee truth. WebAI uses these distinctions to build knowledge that can be challenged and revised.

What you will discover

  • Trace statements back to original texts.
  • Separate attribution, technical authenticity and factual accuracy.
  • Check publication status and demonstration limitations.

Open evidence worksheet ↗

Primary reference points

Links to external institutional sources and official documentation.

W3C — Web & AIModel Context ProtocolAgent2Agent ProtocolC2PA Specifications

No live news collection or AI model connection is enabled in this preview.

DEEPEN YOUR KNOWLEDGE · RESEARCH DOSSIERS

In-depth explanations, concrete cases and their evidence

Every dossier connects mechanisms, cases, limitations and original documents. Independent scientific review remains pending; original sources are provided for verification.

5studies to explore
01 Sécurité

Indirect prompt injection through Web content

Separate authorized instructions from untrusted retrieved Web content.

The trust boundary

A retrieved page, search result or document can embed a hostile instruction targeting the assistant. The model may confuse external text with higher-trust instructions. The attack need not be obvious to a human reading the page.

Beyond visible jailbreaks

Indirect injection becomes especially dangerous when a system can call tools or access private data. A page claiming to contain system instructions has no privilege merely because it was retrieved.

View the verification workflow
  1. Identify untrusted inputs
  2. Constrain tool privileges
  3. Test controlled malicious content
  4. Audit tool calls
Explore explanations, steps and primary sources Read the primary source
02 Confiance

C2PA provenance and independent verification

Avoid conflating signatures, attribution and factual truth.

What is technically signed

A C2PA manifest binds origin or edit assertions about media to signing mechanisms. Validation can confirm integrity of certain claims without proving that the depicted scene corresponds to reality.

What absence means

Missing C2PA metadata does not prove AI generation. Transformations, exports or platforms may strip or fail to carry some information.

View the verification workflow
  1. Find the manifest
  2. Validate available signatures
  3. Inspect declared assertions
  4. Cross-check external evidence
Explore explanations, steps and primary sources Read the primary source
03 Droit et éthique

EU transparency for AI-generated content

Read transparency requirements without generalizing beyond applicable law.

Article 50 and its scope

The EU AI Act provides disclosure duties for certain interactions with AI systems and marking or disclosure for some synthetic content. Obligations, exceptions and application dates need to be read in the consolidated text.

Marking and detection

The law addresses machine-readable marking of certain synthetic outputs to the extent technically feasible. This is different from merely placing a visible graphic watermark.

View the verification workflow
  1. Identify the deployment context
  2. Read the relevant exceptions
  3. Document disclosure mechanisms
  4. Seek legal review
Explore explanations, steps and primary sources Read the primary source
04 Gouvernance

Human oversight of Web agents

Design approval gates before consequential actions.

Distinguish observing from acting

Reading a page or organizing results carries different risks from sending messages, deleting files or completing transactions. Separate read, draft and execute capabilities with different approval conditions.

Meaningful consent

Users should understand the tool identity, action target, shared data and likely consequences before confirming. Invisible blanket permission is not a substitute for informed approval of a specific operation.

View the verification workflow
  1. Classify action risks
  2. Define permissions
  3. Test approval gates
  4. Verify final system state
Explore explanations, steps and primary sources Read the primary source
05 Gouvernance

Assess risks of the generative Web

Structure risk assessment rather than relying on promises.

A framework, not a certificate

The NIST AI RMF and its generative AI profile provide categories and actions for risk management. They are voluntary resources, not automatic legal compliance or proof of reliability.

Identify the actual system

Specify who uses generative Web tools, for which task, with which models, data, permissions and consequences. A draft generator has a different risk profile from an agent executing sensitive operations.

View the verification workflow
  1. Describe the actual use case
  2. Identify potential harms
  3. Test mitigations
  4. Schedule risk review
Explore explanations, steps and primary sources Read the primary source

Experimental proposals, published standards and educational interpretations are distinguished in the dossiers. Research snapshot: 13 November 2026.

Explore the full encyclopedia ↗