Should US Interpreting Companies be Forced to ‘Repatriate’ Overseas Call Centers?

Two US senators have introduced the “Keep Call Centers in America Act of 2025” (S.2495) to the U.S. Senate, a proposed federal legislation that would penalize employers who outsource call center operations and would restrict the use of AI in customer service. 

In the proposed bill, a call center is vaguely defined, potentially encompassing remote interpreters and multilingual customer service agents. If voted into law, S.2495 could significantly impact the language industry, as many large Language Solutions Integrators (LSIs) providers rely on globally distributed staff.

The first part of the bill would require US employers to notify the government before moving call center work overseas, specifically if that line of work represents at least 30% of their annual volume. Companies that outsource the work would be added to a public list and become ineligible for federal grants or guaranteed loans for five years, with few exceptions.

The second key provision of the bill focuses on customer service communications. It would mandate that businesses disclose if AI is being used and that human agents reveal their physical location at the beginning of a call. Consumers would then have the right to request a transfer to a US-based human agent, also with a few exceptions.

When asked their opinion as to whether the US government should regulate call center location, close to half of our readers (47.3%) said it is a great idea. To about 1 in 5 (21.1%), the bill is directionally interesting. The rest are divided among those who think it would probably do more harm than good (15.8%) and those who think the government should keep out (15.8%).

Self-Driven or Copilot[ed]?

Microsoft researchers identified translators and interpreters as frontrunners in the list of occupations whose work aligns most closely with frequent, successful tasks performed by Microsoft’s Copilot.

The findings, based on an analysis of 200,000 US anonymized Copilot user chats against several sources of occupational data, revealed that AI is most often used for gathering information and writing. 

By breaking down professional roles by their associated tasks and comparing them to Copilot’s capabilities, researchers calculated an “AI applicability score,” which indeed was highest for translators and interpreters. 

The study found that 98% of their work activities overlap with tasks where Copilot does well, such as checking information for accuracy and advising on educational topics.

“Our data is only about AI usage and we have no data on the downstream impacts of that usage, so we only weigh in on the automation vs. augmentation question by separately measuring the tasks that AI performs and assists,” stated the researchers.

They also cautioned against assuming that a high applicability score means a profession is at risk of being automated, emphasizing that the data does not account for the broader business impacts of new technology. 

We asked Slator readers if they use Copilot, and over half (52.3%) of respondents have never used it. Less than a quarter (20.0%) rarely use it. About one in six (15.4%) use it often, and the rest (12.3%) use it sometimes.

I’ll Hedge the 5th

OpenAI’s GPT-5 was grown in-lab for over two years, its long-awaited release on August 7, 2025, preceded by an ever-growing list of models and their minis for users to choose from — and quite a bit of hype.

Reactions on social media were mixed, and suggested that the new model may not live up to the hype. And while previous updates, like GPT-4o, featured live demonstrations of capabilities with immediate relevance to the language industry, the focus for GPT-5 was on improvements in areas like reasoning and coding.

No obvious improvement on translation this time: an analysis of GPT-5’s “System Card” revealed that the model’s multilingual capabilities (evaluated by the MMLU benchmark) show little to no improvement over its predecessors. 

OpenAI did highlight voice advancements, but a footnote in its announcement clarifies that the new voice mode is still powered by the older GPT-4o. And while the announcement said the GPT-3 and GPT-4 models would be deprecated, GPT-4o is in fact still available at writing time on the list of model options as a “legacy model.”

We asked readers to share their first impression about GPT-5, and surprisingly, close to half of respondents (46.2%) are not users (yet). A quarter of readers see no obvious improvement (25.6%), while about 1 in 7 (12.8%) see a slight improvement. About a tenth of respondents (10.3%) did see a big improvement, and the rest (5.1%) miss the older models already.

Marketing Confluence and Affluence

There are many ways to market a new AI live speech translation product (and there are so many…). One way is what startup Naitiv.ai chose to do, which was to use its AI tool anonymously on multiple Discord channels to converse in Spanish, Korean, and Japanese, and collect user feedback.

Naitiv’s Founder and CEO, Gayatri Shahane, shared details of their experience during SlatorPod episode #261. It is certainly a creative and cost-effective way to test the technology in a real-world, unscripted environment.

In stark contrast, Google invested some of its substantial marketing budget on late-night TV host Jimmy Fallon to participate in a video promoting its new voice translation feature for Pixel phones. The scripted and rehearsed result can be seen on “Made by Google 25,” streamed live on August 20, 2025, on YouTube.

Speech translation is becoming a feature on many operating systems and platforms, and Meta joined the trend by launching free AI audio translation for creators on Facebook and Instagram. It gives them the option to automatically dub and lip-sync Reels, following closely behind similar updates from YouTube.

We asked readers what they would choose for a marketing campaign, and close to 1 in 2 (42.9%) would pick something in between Fallon and Discord. About a third (33.3%) would go for a Discord group, and the rest (23.8%) for Fallon.