Klarvo AI Transparency for EU AI Act
| 开发者 | klarvo |
|---|---|
| 更新时间 | 2026年6月13日 05:34 |
| PHP版本: | 7.4 及以上 |
| WordPress版本: | 7.0 |
| 版权: | GPLv2 or later |
| 版权网址: | 版权信息 |
详情介绍:
The EU AI Act's transparency rules — Article 50 — start to apply on 2 August 2026. If your website uses AI in any form — a chatbot, AI-written posts, AI personalisation, AI-generated images — your visitors have a right to be told. Most plugins hand you a generic banner and leave the hard part, working out what actually applies to you, on your desk.
Klarvo does the hard part for you. Activate it and, in about a minute, you have a clear answer and the disclosures already in place.
Detection that doesn't miss
Anyone can match your plugin list against a database. The AI that gets a site caught out is the AI a list never sees — and finding it is what Klarvo is built for. It looks four ways at once:
- Your plugins and themes — against a living registry of 1,800+ AI tools: chatbots, content generators, translation, personalisation, search and more.
- Your actual published pages, read server-side — so a chat widget dropped in as a snippet (Tidio, Crisp, Intercom, HubSpot and the like) is caught even when no plugin gives it away.
- What loads in your visitors' browsers, at runtime — so an assistant that only appears once your JavaScript runs is caught too, with nothing to configure.
- AI it has never seen before — with a free account, when your site runs something brand-new or niche that no library has catalogued yet, KlarvoEngine reads that tool's own public description and decides whether it's an AI system you need to disclose. Only the tool's public details are used — never your content, never your visitors' data.
- It finds the AI you're actually running — not just from a list, but by reading your live pages and watching what loads at runtime (and, with a free account, reasoning about tools no library has catalogued yet).
- It gives you one honest answer — an AI Act Readiness score out of 100 and a plain traffic light: Covered, Action needed, or At risk. No jargon, no fear.
- It tells you what that means for you, in plain English, with the exact part of the law one click away.
- It sets your disclosures up for you — a compliant Article 50 notice, an AI usage statement page, a chatbot label, AI-content labelling — then keeps watching, re-checking automatically whenever your site changes.
- Multi-layer detection that doesn't miss. Your plugins and themes, your actual published pages (read server-side), and what loads at runtime — so embedded chat widgets and script-only AI don't slip through. All standalone, no account needed.
- One Compliance Dashboard. Your readiness score, a traffic light, and every finding spelled out: what we found, what the law says, and a button that fixes it.
- A one-minute setup wizard. Auto-detects what it can, asks only what it can't, then creates your disclosures for you — live on your site.
- A compliant Article 50 notice. Accessible, isolated in a Shadow DOM so it never collides with your theme. Light, dark or auto; badge or bar.
- A point-of-interaction chatbot label. A clear "you're talking to AI" cue where the conversation happens (Article 50(1)), with a ready-made greeting snippet.
- AI-content marking. Mark posts or pages as AI-generated, one at a time or in bulk, with the right Article 50 disclosure.
- An AI usage statement page, written for you and tailored to your site.
- An always-on watchtower. Re-checks automatically when a plugin or post changes, and keeps your detection registry current.
- Accessible & private by default. Keyboard-navigable, screen-reader correct, WCAG AA contrast, respects reduced-motion and colour-scheme. No cookies, no tracking — in standalone mode nothing is sent anywhere.
- Law-change radar. A dated, curated view of how the EU AI Act's transparency rules are actually moving — the 2 August 2026 date, the AI-literacy duty already in force, the Code of Practice and guidelines being finalised — flagging what affects your site. A signal, never a scare.
- AI Act Readiness Report. A dated, shareable report (save it as a PDF from your browser) of the AI on your site, your Article 50 obligations, and the disclosures in place. Your evidence that you checked.
- Deepfake & AI-image labelling (Article 50(4)). Mark a realistic AI-generated or manipulated image and Klarvo adds a clear, visible "AI-generated" label wherever it appears — and can preserve Content Credentials (C2PA) when WordPress resizes it.
- AI-literacy pack (Article 4). A short, site-specific briefing for your team plus a simple acknowledgement register — both downloadable as a dated record. A proportionate way for a small team to show it took the duty seriously.
- EU AI Act self-assessment. A few honest questions produce a tailored checklist of what genuinely applies to you, a dated record to keep, and plain-English answers to the things most guidance gets wrong.
- Catch AI no library has seen yet. KlarvoEngine reads the public description of an unrecognised tool and tells you whether it needs an Article 50 disclosure — public metadata only, never your content or your visitors' data.
- Sync detected tools to Klarvo so your full EU AI Act inventory starts already populated.
- Pull app-managed disclosure text straight into your notice.
- Manage multiple sites and keep a durable record from one place.
- There is no small-business exemption from Article 50 — it applies to organisations of every size.
- You don't have to label every AI-assisted blog post — Article 50(4) targets realistic deepfakes and public-interest text published without human review.
- The high-risk rules were deferred, but Article 50 keeps its 2 August 2026 date.
安装:
- Install and activate the plugin.
- Klarvo scans your site immediately and opens a one-minute setup — it sets your disclosures up for you.
- Visit AI Transparency in your admin menu any time to see your readiness score and findings.
- Optional: paste a linking code from your Klarvo account (Integrations → Connect a site) to turn on connected features.
屏幕截图:
常见问题:
Do I really need this before 2 August 2026?
If your site uses AI that visitors interact with or consume — a chatbot, AI-written content, AI-generated images — then Article 50 expects you to disclose it from 2 August 2026. Klarvo finds what applies to you and puts the disclosures in place now, so the date is a non-event.
How does Klarvo find AI that other plugins miss?
Most transparency plugins check your installed plugins against a list and stop there. Klarvo looks four ways at once: your plugins and themes (against a registry of 1,800+ AI tools); your actual published pages, read server-side (so a chat widget added as a snippet — Tidio, Crisp, Intercom, HubSpot — is caught even though no plugin shows it); what loads in the browser at runtime (an assistant that only appears once your JavaScript runs); and, with a free account, AI it has never seen before, where KlarvoEngine reads a tool's own public description and decides whether it needs disclosing. And if it can't fully check your site, it tells you plainly, instead of assuming you're clear.
Is it actually free?
Yes. The scan, the readiness score, the Article 50 notice, the setup wizard, content and image labelling, the readiness report, the literacy pack and the self-assessment all work fully on their own, with no account. A free Klarvo account only adds optional syncing, app-managed text and multi-site management.
My site doesn't use AI. Is this still useful?
Yes — that's exactly when peace of mind matters. Klarvo confirms you're covered, then keeps watching, so if you add a chatbot or AI content later it spots it and tells you what to do. You never have to remember to check.
Does it send my data anywhere?
In standalone mode, no — nothing leaves your site. If you connect a Klarvo account, the plugin talks only to Klarvo's API: to sync your detected tools, pull your disclosure text, and — when it meets a tool it doesn't recognise — to ask KlarvoEngine about that tool using only its public details (its name, description and link), never your content and never your visitors' data. The detection-registry refresh uses a public, anonymous endpoint.
Will it slow my site down?
No. The notice loads after your page (deferred), renders in an isolated Shadow DOM, and adds no layout shift. The scan runs in your admin, never on visitor page loads.
Does installing this make me legally compliant?
No, and we won't pretend otherwise. Klarvo is a technical transparency tool that helps you find and disclose the AI on your site, and keep a dated record. It is not legal advice.
Is the notice accessible?
Yes — keyboard-navigable, screen-reader correct, WCAG AA contrast, and it respects reduced-motion and colour-scheme preferences.
I'm upgrading from 2.x — will my settings survive?
Yes. Your display settings and any AI-content declarations are preserved, and your notice keeps working. You'll land on the new dashboard with your current status, and you can run the setup any time. If you previously connected an account, you'll be asked to reconnect once.
更新日志:
5.1.0
- New: Try Klarvo before you install it. The WordPress.org listing now offers a Live Preview — a staged demo site in WordPress Playground that opens straight on a populated Compliance Dashboard.
- One-time review ask: after your scan completes with a green or amber verdict, the dashboard asks once — and only once, ever — whether Klarvo has earned a review. Dismiss it (or just refresh) and it never returns.
- Housekeeping: uninstall now also removes the newer detection-state options (the server-side fetch result and KlarvoEngine recognitions), keeping "delete removes all plugin data" exactly true.
- Non-destructive upgrade: every setting, declaration, and connection is preserved.
- The visitor notice now explains each disclosure in plain language, per Article 50 clause — "You can interact with this AI system (Article 50(1))" reads differently from "Some content here is AI-generated (Article 50(4))" — in standalone as well as connected mode. All four new lines are translatable.
- The notice script version now matches the plugin version (housekeeping; no behaviour change).
- Non-destructive upgrade: every setting, declaration, and connection is preserved.
- Clearer plugin listing: a plain-English account of how Klarvo's detection actually works — your plugins and themes, your published pages (read server-side), what loads at runtime, and (connected) AI it has never seen before, classified by KlarvoEngine from public metadata. No functional changes; all your settings, declarations and connection are preserved.
- New: Klarvo now recognises AI tools it has never seen before. When you're connected to a free Klarvo account and your site runs a plugin that isn't yet in Klarvo's detection library, Klarvo asks KlarvoEngine to read that plugin's own public description and decide whether it's an AI system that needs disclosing under Article 50 — so even brand-new or niche AI tools get caught and added to your readiness score. Only the plugin's public details are sent (its name, description and link) — never your content or your visitors' data.
- Standalone sites stay completely private and offline: this reasoning runs only with a connected account, and standalone installs simply receive the same recognitions later, free, through the normal detection-library updates.
- Non-destructive upgrade from 4.x: every setting, declaration, the connection, and your AI usage page are preserved.
- New: runtime detection. Some AI tools only appear once a page's JavaScript runs — a chat widget mounted by a script, an assistant exposed as a browser global, with no plugin and no obvious script URL for a server-side scan to catch. Klarvo's on-page notice now reports those back to your readiness score (through your site's own WordPress, never an external service), so they're detected and disclosed like everything else. Safe by design: it only ever recognises tools already in the detection registry, and a report can never make your score more lenient.
- Non-destructive upgrade from 4.2.x: every setting, declaration, the connection, and your AI usage page are preserved.
- Fixed (important): a site whose only AI is a script-embedded chat widget (Tidio, Crisp, Intercom, HubSpot and the like) — added by a snippet rather than a plugin, and not declared — could previously score a misleading 100 / "You're covered." Klarvo now does a server-side check of your own published pages and matches those embedded tools, so they're detected and disclosed like any other. No account or external service needed — it reads your own site, nothing else.
- New: an honest "We couldn't fully verify this site" state. If Klarvo can't read your live pages (a timeout, a login wall, or a cookie-consent tool that holds third-party scripts back), it now says so plainly and offers a one-click re-check — instead of assuming everything's clear. The readiness score never claims a clean bill of health it can't back up.
- Non-destructive upgrade from 4.1.x: every setting, declaration, the connection, and your AI usage page are preserved.
- Polish: the 2 August 2026 Article 50 deadline now appears where it matters — on the dashboard, the "you're covered" states, the setup wizard, and the Readiness Report — so you always know exactly what you're being made ready for. Calm and factual, never alarmist. No functional changes.
- New: Law-change radar — a "rules last reviewed" date and a curated, dated view of how the EU AI Act's transparency rules are developing (Article 50's 2 August 2026 date, the AI literacy duty already in force, the Code of Practice and guidelines being finalised), flagging what's relevant to your site. Honest by design: it's a signal, not legal advice, and it never scaremongers.
- New: AI Act Readiness Report — a dated, shareable report (save it as a PDF from your browser) showing the AI on your site, your Article 50 obligations, and the disclosures in place. Your proof you checked. Not legal advice.
- New: Deepfake / AI-image labelling (Article 50(4)) — mark a realistic AI-generated or manipulated image in your Media Library and Klarvo adds a clear, visible "AI-generated" label wherever it appears. Optionally preserve Content Credentials (C2PA) when WordPress resizes images. Honest about the law: the machine-readable watermark is the AI provider's duty (50(2)); the visible label is yours (50(4)), and you needn't label every AI-touched graphic.
- New: AI-literacy pack (Article 4) — Article 4 is already in force. Klarvo generates a short, site-specific AI-literacy briefing for your team and keeps a simple acknowledgement register, both downloadable as a dated record. A proportionate way for a small team to show it's taken the duty seriously.
- New: EU AI Act self-assessment — a few honest questions (deployer or provider? chatbot? AI text or images?) produce a tailored checklist of what actually applies to you, a dated record you can keep, and plain-English copy that corrects the things most guidance gets wrong (there's no SME exemption from Article 50; the penalty tier is the lower of €15m / 3% for SMEs).
- Non-destructive upgrade from 4.0.x: every setting, declaration, the connection, and your AI usage page are preserved.
- New: Compliance Dashboard — an AI Act Readiness score (0–100) and a Covered / Action needed / At risk traffic light, with plain-English findings and one-click fixes.
- New: Setup wizard that scans your site and sets up your disclosures for you — the Article 50 notice, an AI usage statement page, a chatbot label, and AI-content labelling.
- New: First-class "no AI detected" state — a clear "you're covered" instead of silence, plus an always-on watchtower that re-checks when your site changes.
- New: Point-of-interaction chatbot label (Article 50(1)) and an exact greeting snippet for your bot.
- Improved: honest, misconception-busting guidance and a clear "not legal advice" position throughout.
- Non-destructive upgrade from 2.x/3.x: settings and content declarations are preserved; your notice keeps working.
- Rebuilt for the Klarvo v2 platform and a new, versioned API contract.
- Unified, Shadow-DOM transparency notice shared with the Klarvo embeddable widget — light/dark/auto themes, badge or bar styles, full accessibility.
- Detection registry expanded to 1,800+ AI tools, kept current automatically.
- Connected features now use a secure site token.
- Non-destructive upgrade from 2.x.
- Detection registry and stability improvements.