Path58 AI-Content Disclosure

The posture in one paragraph

Path58 ships content authored primarily with AI assistance — images, audio, video, and text generated, edited, or synthesised with AI models. We declare this transparently as a standing posture rather than labelling each artefact, because we believe a standing declaration plus per-asset machine-readable metadata is the cleaner contract with our audience than per-shipment visible markers that "stain" content. Per-asset attribution (model used, voice attribution, generation date) rides in standard metadata — C2PA Content Credentials for images, ID3 and Vorbis container metadata for audio and video — that any audience member can inspect. Our AI-content generation pipeline enforces this: no asset ships from Path58 without metadata attached.

§ 1 Scope — the Path58 family

This disclosure covers the Path58 family of businesses under common control (Tsvika Vagman, founder and principal): Path58 Consultancy (the revenue/GTM consultancy at path58.com and associated thought-leadership / case-study surfaces) and Path58 Products (the software family — p58-n8n MCP, the competitive-shootout benchmark dashboard, and AX Engine). The same disclosure posture applies across the family. When an AI-generated image, audio clip, video, or AI-assisted text appears on any Path58 surface — under the Path58 consultancy brand, the p58 product brand, or attributed to Tsvika personally — this declaration covers it.

§ 2 What we consider AI-generated content

Three media classes are in scope: image (still or animated; generated by Midjourney, FLUX, DALL-E, Stable Diffusion, Adobe Firefly, or any successor; AI-upscaled or AI-restyled photography that materially alters the source; AI-generated illustrations, diagrams, or carousel visuals); audio (text-to-speech output, AI voice clones including the synthesised clone of Tsvika's voice trained on his real speech, AI-generated music or sound effects); and video (deepfake video, AI-generated avatar video, AI-generated B-roll or motion graphics, hybrid videos combining real footage with AI-generated overlays).

Text is partially in scope. AI-assisted text (LinkedIn posts drafted with AI assistance, blog drafts copy-edited by AI, AI-generated headlines tested in variants) passes through human editorial review under Tsvika's named editorial responsibility before publication. Under EU AI Act Article 50(4), AI-assisted text published under named editorial responsibility is exempt from mandatory disclosure. Our standing posture still covers this content — we declare it openly here rather than relying on the regulatory carve-out.

Standardly retouched photography (colour correction, cropping, exposure), standardly edited audio (noise reduction, level normalisation), and standardly edited video (cuts, colour grading, music underlay) are not in scope — Article 50(2) excludes AI that "performs an assistive function for standard editing or does not substantially alter the input data."

§ 3 How disclosure works — three layers

Disclosure works in three layers, applied together as a single governance system. The first two layers reach the audience; the third layer guarantees the first two are present on every shipment without depending on per-shipment human discipline.

Layer 1 — Standing declarations (this page + per-surface bios and footers)

A persistent declaration is published once per surface and remains live: this page is the canonical version; per-channel bio updates (LinkedIn company-page Overview, Tsvika personal-profile About, X bios, Substack about, Pulse masthead, PDF/deck cover slides, email signatures, page footers across path58.com) carry channel-appropriate versions of the same declaration. A person interacting with any Path58 surface encounters the declaration before or alongside any individual AI-generated artefact. This satisfies EU AI Act Article 50(5) "clear and distinguishable manner at first interaction."

Layer 2 — Per-asset metadata (machine-readable attribution)

Every AI-generated artefact carries the per-format metadata schema appropriate to its media class. For images: C2PA Content Credentials where the vendor supports it, preserved through our publishing pipeline. For audio: container metadata (ID3 for MP3, Vorbis comment for OGG, equivalent for AAC/MP4) with a synthesis-marker tag identifying the vendor and model. For video: container-level metadata with synthesis tags. This satisfies EU AI Act Article 50(2) machine-readable detectability and supplies per-shipment attribution to any audience member who inspects the file. No visible per-asset markers (no on-screen badges, no inline-prose disclosures attached to individual shipments, no audible pre-rolls on synthesised voice) are applied — the standing declaration in Layer 1 is the operative human-perceivable layer; metadata in Layer 2 is the per-asset attribution layer.

Layer 3 — Pipeline enforcement

Our AI-content generation pipeline is mandatory and non-bypassable. Every AI-generation node injects metadata before output. A hard-fail gate refuses to emit any asset whose metadata schema is invalid or missing. A pre-publish gate refuses shipment to any channel if the asset lacks metadata. A weekly post-publish audit cron scans shipped assets across all Path58 surfaces and surfaces any gap to the operator. Manual one-off shipments that bypass the pipeline must be logged in an exception log reviewed weekly. The pipeline guarantees Layers 1 and 2 are present on every shipment without depending on per-shipment human discipline.

§ 4 Regulatory authority — EU AI Act Article 50

Path58 is established in Israel; we ship content to EU audiences via LinkedIn, Substack, this website, and direct outreach to EU contacts. EU AI Act Article 50 attaches to deployers of AI systems whose outputs are made available to natural persons in the Union. Path58 cannot exclude EU exposure without geo-fencing, and we choose not to. We comply with the sub-clauses of Article 50 that bind us:

• Article 50(2) — provider-side machine-readable marking. Path58 is not the provider of the underlying models, but we preserve vendor-produced marking (C2PA, container metadata) through our publishing pipeline rather than stripping it.
• Article 50(4) — deployer-side disclosure for deepfake image/audio/video. The standing declaration above is Path58's operative disclosure. For synthesised audio that resembles Tsvika's voice specifically (the one Article 50(4) "deepfake" definition trigger case in our content), the standing-declaration layer covers the human-perceivable disclosure obligation; container metadata identifies the voice clone per-asset. We accept a small interpretive risk that a regulator could disagree the standing layer is "clear and distinguishable enough" for voice-clone audio surfaced to a person with no prior exposure to this page; if regulator guidance changes or a complaint surfaces, we will revisit per the periodic review cadence below.
• Article 50(4) text exemption — AI-assisted text published under Tsvika's named editorial responsibility is exempt from mandatory disclosure. Our standing declaration covers this content voluntarily.
• Article 50(5) — clear and distinguishable disclosure at first interaction. The standing declaration is published at every Path58 surface a person could plausibly encounter Path58 content from.
• Article 50(1) — AI-as-interlocutor disclosure. Path58 does not currently ship customer-facing AI chatbots; if we do, the surface will disclose AI nature at first interaction.
• Article 50(3) — emotion recognition and biometric categorisation. Not applicable; we deploy none of these systems.

§ 5 Operator-truth posture — why we declare rather than per-item label

Our disclosure posture is governed by the operator-truth doctrine that also governs our Privacy Policy: a small founder-led firm does not commit to promises it cannot keep at every stage. A standing declaration plus pipeline-enforced metadata is a promise we can keep across a normal content cadence; per-item visible labels on every shipment is a discipline that breaks the first time we ship under pressure. We choose the keepable promise. We also choose the more transparent posture — declaring across all our content rather than only the legally-mandated cases — because content un-stained-by-labels-but-disclosed-at-the-source is a cleaner contract with our audience than the per-item-label alternative.

§ 6 Retroactivity discipline

This disclosure applies to AI-generated content created or shipped on or after 2026-06-27 (the effective date above). At ratify, Path58 audited its pre-existing content surfaces and confirmed no in-scope AI-generated image, audio, or video content had been shipped under a prior regime — the disclosure applies cleanly to all AI content from this point forward.

§ 7 Vendor disclosure — current and future

Path58 uses a small set of AI vendors. As of the effective date, we are evaluating image vendors (Midjourney V7, FLUX, OpenAI image models) under our content-system requirements; audio vendor (ElevenLabs Pro Voice Clone, for the synthesised Tsvika voice trained on his real speech); and video vendor (HeyGen, Synthesia for AI-avatar segments in hybrid videos). When a vendor is ratified, this page will list the vendor's identity and the specific model in use, along with the metadata schema applied to that vendor's outputs. Vendor-specific addendums to our internal protocol attach when each vendor ratifies; this page is updated at the periodic review cadence below.

§ 8 Periodic review

EU AI Act Article 50 jurisprudence is expected to evolve through 2027 and beyond. The AI Office is empowered under Article 50(7) to encourage codes of practice for detection and labelling of artificially generated content. We review this disclosure at six-month intervals starting 2027-02-02 — covering new regulatory guidance, vendor-side capability evolution (C2PA support broadening, new synthesis-marker standards), new jurisdictions where Path58 audience exposure has materially grown (FTC AI labeling rules, California AB-3211 or equivalent), and operational learnings. Out-of-cadence reviews trigger on material regulatory events, complaints, court rulings, or vendor changes that affect our implementation.

§ 9 Contact

For any AI-disclosure question, regulator inquiry, or complaint:

Email: [email protected]
Response time: 5 business days is our internal target.
Languages: English or Hebrew

This page is the audience-facing mirror of an operator-ratified internal protocol. The internal protocol contains the full per-channel adoption rules, per-format metadata schemas, pipeline-enforcement specification, and amendment history. The internal protocol is canonical; this page reflects its operative posture. For research or transparency requests beyond what is covered here, contact us at the email above.