LLM Translation enters the mainstream: EU and Google say it's good enough without humans

Two recent signals from the world’s most influential players—the European Union and Google—have redefined how we think about LLM-based translation (or AI translation). Both the EU and Google now accept that AI-powered translation is fluent, usable, and suitable for most everyday applications without penalties, fines, or mandatory human review.

This quiet shift marks a tipping point that we have heard again at again at many industry events: Machine translation is no longer experimental—it’s an accepted, reliable, and scalable communication tool.

EU: LLM Translation is “low risk” and fit for purpose

With the enforcement of the EU AI Act in August 2024, the European Union is officially classifying most translation systems based on artificial intelligence as either “minimal risk” or “low risk.” For practical purposes, this means AI translation is now considered safe for internal documentation, customer support, websites, and a wide range of commercial and administrative tasks—without requiring constant human validation. The EU itself (the world's most multilingual organization employing the largest pool of in-house and freelance translators) is translating most of its website pages using its own neural machine translation system, eTranslation. Pangeanic supported eTranslation on several occasions, not only in its early stages as a back-up system (the iADAATPA Project)but also with data and benchmarking with its NTEU Project, releasing more than 500 neural engines and a deluge of parallel corpora to train AI Systems, which has been further used to train LLMs such as SalamandraTA by the Barcelona Supercomputing Center.

Only high-stakes use cases—such as legal, financial, or medical content—still require a human-in-the-loop approach to ensure that potential errors don’t lead to serious harm. But for the majority of business operations and citizen services, AI translation is not just permissible; it’s encouraged as a scalable solution for Europe’s multilingual reality. This has had a small impact on us at Pangeanic as we have the Spanish Tax Office as a star use case with a private SaaS deployment: we have been asked to watermark all translated documents.

Google: No longer penalizing automated translations

In parallel, Google has revised its stance on automatically translated content. Once wary of machine-translated pages for fear of poor user experience, Google now states that "AI-generated translations are acceptable—as long as they provide value to users and comply with its general content quality guidelines".

What changed? Google recognizes that AI translation has significantly improved in quality, especially with the advent of context-aware systems. Instead of penalizing content based on how it was created (human vs. machine), Google now evaluates its utility, readability, and relevance... and it doesn't matter whether the creator was a human, or a human controlling a prompt, or an automated workflow (surely not the best source for high-quality experiences).

This policy change brings automated translation fully into the realm of SEO-friendly content, unlocking its potential for global content strategies, localization workflows, and the generation of long-tail content.

.

What the EU and Google now agree on

Across both institutions, a common conclusion is now clear:

AI translation is “good enough” for most real-world use—and increasingly essential.

So, the European Union, through its AI Act in 2024, and Google, through its search and content indexing policies in 2025, have both arrived at a shared position that affirms the maturity and fluency of modern machine translation systems. Their respective decisions—regulatory in one case, algorithmic in the other—represent a societal shift in how AI translation is perceived, regulated, and operationalized.

We now two massive reference points for wide adoption of machine translation at scale:

• AI translation is safe and effective for large-scale communication.
  Whether it's internal knowledge-sharing, cross-border customer service, or public-facing content, MT can now be deployed confidently without the need for exhaustive human oversight. This reduces cost, accelerates workflows, and enables true multilingual scale.

• Human review is only needed where the stakes demand it.
  And this is decreasing and the role is changing. In 2016, CSA Research conducted a thought experiment that found that translation firms translate just 0.00000000009% of content generated each day. "In 2019, Google alone translated 300 trillion words compared to an estimated 200 billion words translated by the professional translation industry" said TAUS founder Jaap van der Meer in a Multilingual magazine article in 2021, forecasting higher and higher demand in the 2020s. And so it has happened. In critical domains like healthcare, legal contracts, or financial disclosures, human validation remains necessary—and rightly so although paradoxically. But for the overwhelming majority of use cases, AI translation delivers sufficient quality to stand on its own (more about this below). This pragmatic differentiation allows organizations to focus their linguistic resources where they matter most.

• Machine translations promote inclusivity and accessibility across languages.
  By lowering the barriers to publishing in multiple languages, AI translation acts as an equalizer, giving SMEs, NGOs,  educational institutions, and underrepresented communities access to tools once reserved for major players with deep localization budgets. The EU views this as part of its broader digital inclusion strategy, while Google sees it as a means to enhance the global user experience.

What’s remarkable is not only the shared conclusion, but how complementary the EU and Google’s approaches are. One governs the use of AI; the other governs its visibility. This growing consensus is not a simple regulatory shift; it reflects a deepening trust in AI’s capacity to bridge language gaps responsibly, efficiently, and at scale. It sets the foundation for a new standard in global communication—one where speed, cost efficiency, and linguistic reach are no longer trade-offs, but simultaneous outcomes. Translation is becoming a task, an agentic task if you ask me.

At Pangeanic, this evolution affirms what we’ve long championed: that AI translation, when developed ethically and deployed intelligently, can empower businesses and institutions to speak the language of their audiences—securely, fluently, and at scale. Our investment in Deep Adaptive MT, anonymization frameworks, and enterprise-grade workflows anticipates exactly this convergence of policy and practice.

The historic double standards in translation are beginning to be over

A 2023 internal quality assessment of one million randomly selected human-translated words, reviewed by professional linguists, found an average of one error per 150 words. These errors include issues with terminology, accuracy, style, and language. Even the most respected human translators, considered the industry's gold standard, make measurable mistakes.

AI translation tools, on the other hand, have achieved remarkable accuracy and can process text much faster than humans. Yet, they have often been held to a higher standard because language was considered exclusively human. We trust autopilots when we fly and we trust them for thousands of miles. Most fatal car accidents have humans as the real culprits of the casualties (because they drink too much, they don't wear a seat belt, because they get distracted...) Many expect machine translations to be perfect, but machine translation already surpasses human performance in consistency and efficiency.

So, how could anyone in 2025 judge LLM translation more harshly than human translators?

The Implications: From Compliance Burden to Strategic Advantage

The shift from viewing AI translation as a liability to recognizing it as a low-risk enabler transforms how businesses approach language. Until recently, many organizations saw MT as a risky shortcut or compliance headache. Now, both regulation and platform policies permit the use of AI translation at scale—for knowledge bases, driver-customer communication via an app, selecting food for delivery in a foreign country, customer support, product documentation, and even marketing content—without relying on human post-editing.

This opens the door for automation across entire departments and industries. AI translation becomes a strategic asset, allowing companies to reduce costs, shorten time-to-market, and deliver content in dozens of languages without overextending human resources.

The Post-Post-Editing Era Begins

We are entering a new phase in the language industry, where machine translation no longer needs to be hidden, post-edited, or apologized for. Instead, it’s seen for what it is: a viable, reliable tool for multilingual communication at scale.

This doesn’t make professional human translators obsolete. On the contrary, it frees them to focus on high-value tasks—such as transcreation, quality assurance, and content that truly demands a human touch. But for the bulk of enterprise content, AI translation is ready, usable, and finally recognized as such by the world’s most influential gatekeepers.

Pangeanic is proud to be leading this transition. From powering multilingual EU initiatives to delivering real-time, secure translation services for global clients, we stand at the frontier of trustworthy AI for language.