Anthropic’s new plug-ins for Cowork announced on Friday are sparking jitters in the markets with software, professional services and analytics companies seeing the largest sell-offs.

Last month, Anthropic launched its Cowork model, a “simpler version of Claude Code” prompting concerns among those heavily invested in software companies. Friday’s (30 January) launch of new plug-ins seems to have accelerated the concerns.

This week has seen a strong sell-off in US and European software, professional services and data analytics companies, with the trend continuing yesterday (3 February) and contagion in Asian markets. Commentators are blaming the release of Anthropic’s plugins for Cowork which the AI player says will automate tasks across legal, sales, marketing and data analysis.

The legal space is where organisations like Thomson Reuters makes much of its revenue, so it was one of the players to see an 18pc slump in its share price yesterday, according to Reuters itself, which added that its shares are now down 33pc just this year, having dropped by 22pc in 2025, as fears rise around AI disruption in the legal sector.

Other providers of legal analytics also dropped with the UK’s RELX falling 14pc and Dutch company Wolters Kluwer seeing a drop of 13pc.

And the contagion spread to other software companies and the broader market as AI fuels concerns among investors who are struggling to figure out who the winners and losers will be in the current AI-fuelled economy. According to Bloomberg, a Goldman Sachs basket of US software stocks fell 6pc yesterday – its sharpest one-day drop since the sell-off that followed the initial US tariffs announcements in April.

When Anthropic launched Cowork on 12 January, it described it as a simpler version of Claude Code for non-coding related tasks. It said this new model has more agency – it can read, edit and re-organise files, taking on many of same tasks Claude Code can, but in a more “approachable” form.

Cowork seems firmly targeted at the enterprise market with its promise to make using Claude “for work” easier. Now, the new sector-specific plugins are seen as a particular threat to existing analytics players.

Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.

Russian-state hackers wasted no time exploiting a critical Microsoft Office vulnerability that allowed them to compromise the devices inside diplomatic, maritime, and transport organizations in more than half a dozen countries, researchers said Wednesday.

The threat group, tracked under names including APT28, Fancy Bear, Sednit, Forest Blizzard, and Sofacy, pounced on the vulnerability, tracked as CVE-2026-21509, less than 48 hours after Microsoft released an urgent, unscheduled security update late last month, the researchers said. After reverse-engineering the patch, group members wrote an advanced exploit that installed one of two never-before-seen backdoor implants.

Stealth, speed, and precision

The entire campaign was designed to make the compromise undetectable to endpoint protection. Besides being novel, the exploits and payloads were encrypted and ran in memory, making their malice hard to spot. The initial infection vector came from previously compromised government accounts from multiple countries and were likely familiar to the targeted email holders. Command and control channels were hosted in legitimate cloud services that are typically allow-listed inside sensitive networks.

“The use of CVE-2026-21509 demonstrates how quickly state-aligned actors can weaponize new vulnerabilities, shrinking the window for defenders to patch critical systems,” the researchers, with security firm Trellix, wrote. “The campaign’s modular infection chain—from initial phish to in-memory backdoor to secondary implants was carefully designed to leverage trusted channels (HTTPS to cloud services, legitimate email flows) and fileless techniques to hide in plain sight.”

The 72-hour spear phishing campaign began January 28 and delivered at least 29 distinct email lures to organizations in nine countries, primarily in Eastern Europe. Trellix named eight of them: Poland, Slovenia, Turkey, Greece, the UAE, Ukraine, Romania, and Bolivia. Organizations targeted were defense ministries (40 percent), transportation/logistics operators (35 percent), and diplomatic entities (25 percent).

There is concern that subscribers might be negatively affected if Netflix acquires Warner Bros. Discovery’s streaming and movie studios businesses. One of the biggest fears is that the merger would lead to higher prices due to less competition for Netflix.

During a US Senate hearing Tuesday, Netflix co-CEO Ted Sarandos suggested that the merger would have an opposite effect.

Sarandos was speaking at a hearing held by the US Senate Judiciary Committee’s Subcommittee on Antitrust, Competition Policy, and Consumer Rights, “Examining the Competitive Impact of the Proposed Netflix-Warner Brothers Transaction.”

Sarandos aimed to convince the subcommittee that Netflix wouldn’t become a monopoly in streaming or in movie and TV production if regulators allowed its acquisition to close. Netflix is the largest subscription video-on-demand provider by subscribers (301.63 million as of January 2025), and Warner Bros. Discovery is the third (128 million streaming subscribers, including users of HBO Max and, to a smaller degree, Discovery+).

Speaking at the hearing, Sarandos said: “Netflix and Warner Bros. both have streaming services, but they are very complementary. In fact, 80 percent of HBO Max subscribers also subscribe to Netflix. We will give consumers more content for less.”

During the hearing, Democratic senator Amy Klobuchar of Minnesota asked Sarandos how Netflix can ensure that streaming remains “affordable” after a merger, especially after Netflix issued a price hike in January 2025 despite adding more subscribers.

Sarandos said the streaming industry is still competitive. The executive claimed that previous Netflix price hikes have come with “a lot more value” for subscribers.

“We are a one-click cancel, so if the consumer says, ‘That’s too much for what I’m getting,’ they can cancel with one click,” Sarandos said.

When pressed further on pricing, the executive argued that the merger doesn’t pose “any concentration risk” and that Netflix is working with the US Department of Justice on potential guardrails against more price hikes.

Sarandos claimed that the merger would “create more value for consumers.” However, his idea of value isn’t just about how much subscribers pay to stream but about content quality. By his calculations, which he provided without further details, Netflix subscribers spend an average of 35 cents per hour of content watched, compared to 90 cents for Paramount+.

The Netflix stat is similar to one provided by MoffettNathanson in January 2025, finding that in the prior quarter, on average, Netflix generated 34 cents in subscription fees per hour of content viewed per subscriber. At the time, the research firm said Paramount+ made an average of 76 cents per hour of content viewed per subscriber.

Downplaying Monopoly Concerns

Netflix views Warner as “both a competitor and a supplier,” Sarandos said when subcommittee chair Republican senator Mike Lee of Utah asked why Netflix wants to buy WB’s film studios, per Variety. The streaming executive claimed that Netflix’s “history is about adding more and more” content and choice.

During the hearing, Sarandos argued that streaming is a competitive business and pointed to Google, Apple, and Amazon as “deep-pocketed tech companies trying to run away with the TV business.” He tried to downplay concerns that Netflix could become a monopoly by emphasizing YouTube’s high TV viewership. Nielsen’s The Gauge tracker shows which platforms Americans use most when using their TVs (as opposed to laptops, tablets, or other devices). In December, it said that YouTube, not including YouTube TV, had more TV viewership (12.7 percent) than any other streaming video-on-demand service, including second-place Netflix (9 percent). Sarandos claimed that Netflix would have 21 percent of the streaming market if it merged with HBO Max.

Without meaningful deterrents, Big Tech companies will do what’s profitable, regardless of the cost to consumers. But a new bipartisan bill could add a check that would make them think twice, at least in one area. On Wednesday, Senators Ruben Gallego (D-AZ) and Bernie Moreno (R-OH) introduced legislation that would require social platforms to crack down on scam ads.

The Safeguarding Consumers from Advertising Misconduct (SCAM) Act would require platforms to take reasonable steps to prevent fraudulent or deceptive ads that they profit from. If they don’t, the Federal Trade Commission (FTC) and state attorneys general could take civil legal action against them.

The backdrop to the SCAM Act is a Reuters report from last November. Meta reportedly estimated that up to 10 percent of its 2024 revenue came from scam ads. The company is said to have calculated that as much as $16 billion of its revenue that year was from scams, including “fraudulent e-commerce and investment schemes, illegal online casinos and the sale of banned medical products.”

Making matters worse, Meta reportedly refused to block small fraudsters until their ads were flagged at least eight times. Meanwhile, bigger spenders were said to have accrued at least 500 strikes without being removed. Executives reportedly wrestled with how to get the problem under control — but only without affecting the company’s bottom line. At one point, managers were told not to take any action that could cost Meta more than 0.15 percent of its total revenue. (See what I mean about needing meaningful deterrents?)

According to the FTC, Americans’ estimated total loss from fraud in 2024 (adjusted for underreporting) was nearly $19 billion. An estimated $81.5 billion of that came from seniors.

“If a company is making money from running ads on their site, it has a responsibility to make sure those ads aren’t fraudulent,” Sen. Gallego said in a statement. “This bipartisan bill will hold social media companies accountable and protect consumers’ money online.”

“It is critical that we protect American consumers from deceptive ads and shameless fraudsters who make millions taking advantage of legal loopholes,” Moreno added. “We can’t sit by while social media companies have business models that knowingly enable scams that target the American people.”

The company expects AVs to unlock a ‘multitrillion-dollar’ opportunity.

Uber said it will roll out autonomous vehicles (AV) in London, Madrid, Munich, Hong Kong and a number of US cities, including in California, as the ride-hailing platform eyes leadership in robotaxi services by 2029.

The announcement came alongside a reasonably strong quarter from the company. Revenue for Q4 2025 grew by 20pc to $14.4bn – though short of analyst expectations – while its consumer base grew to more than 200m monthly users completing more than 40m trips daily.

“We enter 2026 with a rapidly growing topline, significant cash flow, and a clear path to becoming the largest facilitator of AV trips in the world,” said Uber CEO Dara Khosrowshahi. The company expects AVs to unlock a “multitrillion-dollar” opportunity.

Uber currently operates its AVs in Abu Dhabi, Dubai and Riyadh, alongside US cities Atlanta, Austin and Dallas. It reported that AV operations in Austin and Atlanta are among its fastest-growing areas in the US.

The expansion marks Uber’s full entry into Europe and Asia with its self-driving taxis. An Uber spokesperson told Bloomberg that the company will partner with previously announced technology providers.

Uber has existing partnerships with the US’s May Mobility, Lucid, and Nuro, China’s Baidu and WeRide, and the UK’s Wayve to test and deploy AVs across metropolitan areas worldwide. It also works with Nvidia – which recently unveiled open-source AI models for self-driving vehicles – to develop the tech behind its robotaxis.

Uber benefits from its established identity as a ride-hailing service provider, but it faces competition in the AV space. Earlier this week, self-driving car company Waymo announced a $16bn funding round, taking it to a $126bn valuation. It plans to grow its services within the US and expand internationally to 20 new cities, including London and Tokyo.

This came after a San Francisco power cut last December shut down Waymo cars in the city, causing gridlock.

Meanwhile, Tesla reported a quarterly revenue drop of 3pc, the first time it has ever reported a revenue decline. The company has been slow to experiment with self-driving taxis, launching a pilot service in Austin last year complete with human safety supervisors. Last month, the company said that it had removed some of the supervisors from its Austin fleet.

Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.

MVK Chari, a pioneer in finite element field computation, died on 3 December. The IEEE Life Fellow was 97.

Chari developed a finite element method (FEM) for analyzing nonlinear electromagnetic fields—which is crucial for the design of electric machines. The technique is used to obtain approximate solutions to complex engineering and mathematical problems. It involves dividing a complicated object or system into smaller, more manageable parts, known as finite elements, according to Fictiv.

As an engineer and technical leader at General Electric in Niskayuna, N.Y., Chari used the tool to analyze large turbogenerators for end region analysis, starting with 2D and expanding its use over time to quasi-2D and 3D.

During his 25 years at GE, he established a team that was developing finite element analysis (FEA) tools for a variety of applications across the company. They ranged from small motors to large MRI magnets.

Chari received the 1993 IEEE Nikola Tesla Award for “pioneering contributions to finite element computations of nonlinear electromagnetic fields for design and analysis of electric machinery.”

A career spanning industry and academia

Chari attended Imperial College London to pursue a master’s degree in electrical engineering. There he met Peter P. Silvester, a visiting professor of electrical engineering. Silvester, a professor at McGill University in Montreal, was a pioneer in understanding numerical analysis of electromagnetic fields.

After Chari graduated in 1968, he joined Silvester at McGill as a doctoral student, applying FEM to solve electromagnetic field problems. Silvester applied the method to waveguides, while Chari applied it to saturated magnetic fields.

Chari joined GE in 1970 after earning his Ph.D. in electrical engineering. He climbed the leadership ladder and was a manager of the company’s electromagnetics division when he left in 1995. He joined Rensselaer Polytechnic Institute in Troy, N.Y., as a visiting research and adjunct professor in its electrical, computer, and systems engineering department. Chari taught graduate and undergraduate classes in electric power engineering and mentored many master’s and doctoral students. His strength was nurturing young engineers.

He also conducted research on electric machines and transformers for the Electric Power Research Institute and the U.S. Department of Energy.

In 2008 Chari joined Magsoft Corp., in Clifton Park, N.Y., and conducted advanced work on specialized software for the U.S. Navy until his retirement in 2016.

Remembering a friend

Chari successfully nominated one of us (Hoole) to be elevated to IEEE Fellow at the age of 40. He helped launch Haran’s career when Chari sent his résumé to GE hiring managers for a position in its applied superconductivity lab.

Chari’s commitment to people came from his family background. His father—M.A. Ayyangar—was known throughout India as a freedom fighter, mathematician, and eventually the speaker of the Indian Parliament’s lower house under Prime Minister Nehru. Chari’s wife, Padma, was a physician in New York.

From Chari’s illustrious family, he was at the peak of South India (Tamil) society.

Chari would fondly and cheerfully tell us the story behind his name. Around the time of his birth, it was common in Tamil society not to have formal names. He went by the informal “house name” Kannah (a term of endearment for Krishna). When it was time for Chari to start school, an auspicious uncle enrolled him. But Chari had no formal name, so the uncle took it upon himself to give him one. He asked Chari if he would like a long or short name, to which he said long. So the uncle named him Madabushi Venkadamachari.

When Chari moved to North America, he shortened his name to Madabushi V.K.

He could also laugh at himself.

A stellar scientist, he also was a role model, guide, and friend to many of us. We thank God for him.

Changes — and subsequent confusion and concern — largely defined the early childhood education sector in 2025. Multiple social programs including Head Start and hunger assistance programs were in flux. Rising costs of living were coupled with the rising costs of child care. And many EdSurge readers were left searching for answers, as seen in our most-read stories of the year.

There was also plenty of innovation in the field, from transforming empty school buildings, adding apprenticeship programs and introducing play into teaching math. There will be more of that undoubtedly in 2026 and EdSurge aims to bring you more answers as questions continue to arise about the future of early learning and child care.

Here are the most popular early childhood education stories, in descending order. You can see our most-read stories covering the K-12 sector here.

10. More Than Half of Child Care Providers Have Gone Hungry, New Report Finds

By Lauren Coffey

Child care providers struggling is nothing new, and many left the field postpandemic due to its low pay and long, unstable hours. But the struggle to survive came to a head last year, as the cost of living continued to rise and multiple social programs — namely SNAP, formerly known as food stamps — were temporarily paused. A report from the RAPID Survey Project at the Stanford Center on Early Childhood found that basic needs may be greater than ever, with 58 percent of child care providers stating they experienced hunger in 2025.

9. Could Play Boost Students’ Math Performance?

By Daniel Mollenkamp

Early education often conjures images of games, bright colors and plenty of play time. But often those associations stop when it comes to math class. EdSurge spoke with experts across the nation looking to marry the two. But similar to the curriculum at ever-popular Montessori schools, “play” is not a free-for-all. When it comes to math instruction, there is a fine line between board and dice games and lessons about larger concepts.

8. What Will Kids Lose If PBS Gets Cut?

By Lauren Coffey

Calls to cut funding for PBS began in the spring of 2025, culminating in multiple slashed grants that more than likely spelled the end for many local public broadcasting affiliates. The cut goes beyond easily accessing beloved shows like “Daniel the Tiger” and “Arthur.” Many experts voiced concerns that the loss of programming, which puts educational guidelines at the forefront, could hit rural and lower-income families particularly hard.

7. As Apprenticeships Expand in Early Childhood, These States Are Training the Field’s Future Leaders

By Emily Tate Sullivan

The leap between early childhood educator and director of an early child care center is often so intimidating that many educators do not attempt to move up, despite it often providing better pay and hours. Registered apprenticeship programs began booming to fix that gap, offering a pathway to train educators for leadership roles. Notably, Kentucky, Massachusetts and New Hampshire offer programs specifically made for emerging leaders in the early education field — and the impact is already being seen.

6. What Will Districts Do With All Those Empty School Buildings? Some Look to Fill Them With Younger Kids

By Emily Tate Sullivan

Enrollment continues to decline in traditional public schools, due in part to the rise of popularity in virtual schools and charter schools buoyed by voucher programs. The outcome: a lot of large, empty school buildings. But some districts, like in Oklahoma City and Tucson, are overhauling them to house early learning programs instead. What follows is a way to address the rising need for child care and a way to lure in early childhood educators, thanks to district benefits.

5. Head Start’s Future Is Uncertain. Rural Americans Aren’t Ready for What Happens Next.

By Claire Woodcock

As the Head Start program turned 60 in 2025, questions swirled about its future. The program, which has long helped families living at or below the poverty level access affordable child care and services, saw half of its regional offices close this year. For most of the year, the fate of its funding was unknown. While Head Start funding was later approved, there was no increase from previous years — bringing concern from many. There is a particular worry about the consequences for rural communities, where 1 in 3 child care programs is backed by Head Start.

4. Study: Kids Suffer as Nearly Half of U.S. Families Struggle to Meet Basic Needs

By Marianna McMurdock

Similarly to our No. 8 story of the year focusing on child care providers, families themselves also struggled this year to make ends meet. A report showed 4 in 10 families are experiencing material hardship, which goes beyond short-term stress: It can hurt children’s learning long-term. Parents’ stress can seep to their children, causing depression and anxiety. It can also cause an overreliance on screen time. The result: children can have a learning gap of up to a year compared to those not experiencing hardship.

3. Why the Dire State of the Early Learning Workforce Is ‘Alarming and Not Surprising’

By Emily Tate Sullivan

Rising costs, staff shortages and low morale brought the early childhood educator crisis to a head in 2025. According to a report by the National Association for the Education of Young Children, high rents and an uptick in property and liability insurance rates has caused stagnant or low revenue for providers, prompting many programs to shutter. Those working in the early childhood world are not surprised by these findings, but do believe more funding and action — versus inaction — is needed.

2. Idaho Moves to Deregulate Child Care in First-of-Its-Kind Legislation

By Emily Tate Sullivan

Idaho made major waves at the start of the year when it attempted to become the first in the nation to eliminate state-mandated child-to-teacher ratios, in a move it believed would help the severe shortage of child care openings. Many experts were quick to defend the ratios as essential to helping with the health of children and the quality of child care. The amended bill ultimately tweaked the ratio proposals, loosening, versus ridding, the requirements.

1. Why Don’t Early Childhood Programs Have Access to Substitute Teachers?

By Emily Tate Sullivan

As winter swings on, bringing with it inevitable sickness, the K-12 system can rely on its large infrastructure of substitute teachers, but the early childhood sector has no such programming. Beyond cold and flu season, this makes it difficult for the already-burned-out teachers in early learning to take a sick day or vacation. However, there are some efforts under way, with many turning toward future full-time educators to fill the gap.

You may see some of my bylines above, and you’ll be seeing more of those in 2026 as I cover more early childhood education for EdSurge. If you have any tips, or just want to say hello, feel free to shoot me a note at lauren@edsurge.com.

Mistral AI, the Paris-based startup positioning itself as Europe’s answer to OpenAI, released a pair of speech-to-text models on Wednesday that the company says can transcribe audio faster, more accurately, and far more cheaply than anything else on the market — all while running entirely on a smartphone or laptop.

The announcement marks the latest salvo in an increasingly competitive battle over voice AI, a technology that enterprise customers see as essential for everything from automated customer service to real-time translation. But unlike offerings from American tech giants, Mistral’s new Voxtral Transcribe 2 models are designed to process sensitive audio without ever transmitting it to remote servers — a feature that could prove decisive for companies in regulated industries like healthcare, finance, and defense.

“You’d like your voice and the transcription of your voice to stay close to where you are, meaning you want it to happen on device—on a laptop, a phone, or a smartwatch,” Pierre Stock, Mistral’s vice president of science operations, said in an interview with VentureBeat. “We make that possible because the model is only 4 billion parameters. It’s small enough to fit almost anywhere.”

Mistral splits its new AI transcription technology into batch processing and real-time applications

Mistral released two distinct models under the Voxtral Transcribe 2 banner, each engineered for different use cases.

• Voxtral Mini Transcribe V2 handles batch transcription, processing pre-recorded audio files in bulk. The company says it achieves the lowest word error rate of any transcription service and is available via API at $0.003 per minute, roughly one-fifth the price of major competitors. The model supports 13 languages, including English, Mandarin Chinese, Japanese, Arabic, Hindi, and several European languages.

• Voxtral Realtime, as its name suggests, processes live audio with a latency that can be configured down to 200 milliseconds — the blink of an eye. Mistral claims this is a breakthrough for applications where even a two-second delay proves unacceptable: live subtitling, voice agents, and real-time customer service augmentation.

The Realtime model ships under an Apache 2.0 open-source license, meaning developers can download the model weights from Hugging Face, modify them, and deploy them without paying Mistral a licensing fee. For companies that prefer not to run their own infrastructure, API access costs $0.006 per minute.

Stock said Mistral is betting on the open-source community to expand the model’s reach. “The open-source community is very imaginative when it comes to applications,” he said. “We’re excited to see what they’re going to do.”

Why on-device AI processing matters for enterprises handling sensitive data

The decision to engineer models small enough to run locally reflects a calculation about where the enterprise market is heading. As companies integrate AI into ever more sensitive workflows — transcribing medical consultations, financial advisory calls, legal depositions — the question of where that data travels has become a dealbreaker.

Stock painted a vivid picture of the problem during his interview. Current note-taking applications with audio capabilities, he explained, often pick up ambient noise in problematic ways: “It might pick up the lyrics of the music in the background. It might pick up another conversation. It might hallucinate from a background noise.”

Mistral invested heavily in training data curation and model architecture to address these issues. “All of that, we spend a lot of time ironing out the data and the way we train the model to robustify it,” Stock said.

The company also added enterprise-specific features that its American competitors have been slower to implement. Context biasing allows customers to upload a list of specialized terminology — medical jargon, proprietary product names, industry acronyms — and the model will automatically favor those terms when transcribing ambiguous audio. Unlike fine-tuning, which requires retraining the model, context biasing works through a simple API parameter.

“You only need a text list,” Stock explained. “And then the model will automatically bias the transcription toward these acronyms or these weird words. And it’s zero shots, no need for retraining, no need for weird stuff.”

From factory floors to call centers, Mistral targets high-noise industrial environments

Stock described two scenarios that capture how Mistral envisions the technology being deployed.

The first involves industrial auditing. Imagine technicians walking through a manufacturing facility, inspecting heavy machinery while shouting observations over the din of factory noise. “In the end, imagine like a perfect timestamped notes identifying who said what — so diarization — while being super robust,” Stock said. The challenge is handling what he called “weird technical language that no one is able to spell except these people.”

The second scenario targets customer service operations. When a caller contacts a support center, Voxtral Realtime can transcribe the conversation in real time, feeding text to backend systems that pull up relevant customer records before the caller finishes explaining the problem.

“The status will appear for the operator on the screen before the customer stops the sentence and stops complaining,” Stock explained. “Which means you can just interact and say, ‘Okay, I can see the status. Let me correct the address and send back the shipment.’”

He estimated this could reduce typical customer service interactions from multiple back-and-forth exchanges to just two interactions: the customer explains the problem, and the agent resolves it immediately.

Real-time translation across languages could arrive by the end of 2026

For all the focus on transcription, Stock made clear that Mistral views these models as foundational technology for a more ambitious goal: real-time speech-to-speech translation that feels natural.

“Maybe the end goal application and what the model is laying the groundwork for is live translation,” he said. “I speak French, you speak English. It’s key to have minimal latency, because otherwise you don’t build empathy. Your face is not out of sync with what you said one second ago.”

That goal puts Mistral in direct competition with Apple and Google, both of which have been racing to solve the same problem. Google’s latest translation model operates at a two-second delay — ten times slower than what Mistral claims for Voxtral Realtime.

Mistral positions itself as the privacy-first alternative for enterprise customers

Mistral occupies an unusual position in the AI landscape. Founded in 2023 by alumni of Meta and Google DeepMind, the company has raised over $2 billion and now carries a valuation of approximately $13.6 billion. Yet it operates with a fraction of the compute resources available to American hyperscalers — and has built its strategy around efficiency rather than brute force.

“The models we release are enterprise grade, industry leading, efficient — in particular, in terms of cost — can be embedded into the edge, unlocks privacy, unlocks control, transparency,” Stock said.

That approach has resonated particularly with European customers wary of dependence on American technology. In January, France’s Ministry of the Armed Forces signed a framework agreement giving the country’s military access to Mistral’s AI models—a deal that explicitly requires deployment on French-controlled infrastructure.

Data privacy remains one of the biggest barriers to voice AI adoption in the enterprise. For companies in sensitive industries — finance, manufacturing, healthcare, insurance — sending audio data to external cloud servers is often a non-starter. The information needs to stay either on the device itself or within the company’s own infrastructure.

Mistral faces stiff competition from OpenAI, Google, and a rising China

The transcription market has grown fiercely competitive. OpenAI’s Whisper model has become something of an industry standard, available both through API and as downloadable open-source weights. Google, Amazon, and Microsoft all offer enterprise-grade speech services. Specialized players like Assembly AI and Deepgram have built substantial businesses serving developers who need reliable, scalable transcription.

Mistral claims its new models outperform all of them on accuracy benchmarks while undercutting them on price. “We are better than them on the benchmarks,” Stock said. Independent verification of those claims will take time, but the company points to performance on FLEURS, a widely used multilingual speech benchmark, where Voxtral models achieve word error rates competitive with or superior to alternatives from OpenAI and Google.

Perhaps more significantly, Mistral’s CEO Arthur Mensch has warned that American AI companies face pressure from an unexpected direction. Speaking at the World Economic Forum in Davos last month, Mensch dismissed the notion that Chinese AI lags behind the West as “a fairy tale.”

“The capabilities of China’s open-source technology is probably stressing the CEOs in the US,” he said.

The French startup bets that trust will determine the winner in enterprise voice AI

Stock predicted that 2026 would be “the year of note-taking” — the moment when AI transcription becomes reliable enough that users trust it completely.

“You need to trust the model, and the model basically cannot make any mistake, otherwise you would just lose trust in the product and stop using it,” he said. “The threshold is super, super hard.”

Whether Mistral has crossed that threshold remains to be seen. Enterprise customers will be the ultimate judges, and they tend to move slowly, testing claims against reality before committing budgets and workflows to new technology. The audio playground in Mistral Studio, where developers can test Voxtral Transcribe 2 with their own files, went live today.

But Stock’s broader argument deserves attention. In a market where American giants compete by throwing billions of dollars at ever-larger models, Mistral is making a different wager: that in the age of AI, smaller and local might beat bigger and distant. For the executives who spend their days worrying about data sovereignty, regulatory compliance, and vendor lock-in, that pitch may prove more compelling than any benchmark.

The race to dominate enterprise voice AI is no longer just about who builds the most powerful model. It’s about who builds the model you’re willing to let listen.