Governments around the world have been struggling to address the rise of industrial-scale scamming operations based in countries like Laos, Myanmar, and Cambodia that have cost victims billions of dollars over the past few years. The operations often have ties to Chinese organized crime, use forced labor to carry out the actual scamming, and rely on vast money laundering networks to collect a profit. They have become so widespread and ingrained in the region that even major international law enforcement collaborations targeting individual scam centers or kingpins haven’t been able to stem the tide.

The FBI said this week that “cyber-enabled” scam complaints from Americans totaled more than $17.7 billion in reported losses last year—likely a major undercount of the real total, given that many victims don’t report their experiences. Some US officials say that a major barrier to comprehensively addressing the issue is the lack of collaboration with Chinese authorities. China’s efforts to address industrial scamming, they argue, appear aimed at reducing the number of Chinese citizens being impacted rather than comprehensively stopping the activity to protect all victims around the world.

“To its credit, China has cracked down on these operations, but it has done so selectively, largely turning a blind eye to scam centers victimizing foreigners,” Reva Price, a member of the US-China Economic and Security Review Commission said at a Senate hearing last month. “As a result, the Chinese criminal syndicates have been incentivized to shift toward targeting Americans.”

According to research the commission published in March, Beijing’s selective strategy has helped embolden some Chinese scammers, even those working within China, to continue operating so long as they exclusively target foreigners.

Other US-based researchers have come to similar conclusions. From 2023 to 2024, China reported a 30 percent decrease in the amount of money its citizens lost to scams, while the US suffered a more than 40 percent increase, according to congressional testimony last year by Jason Tower, who was then the Myanmar country director for the US Institute of Peace’s Program on Transnational Crime and Security in Southeast Asia. In response to Beijing’s enforcement dynamics, Tower said at the time, “the scam syndicates are increasingly pivoting to target the rest of the world, and especially Americans.”

The United Nations Office on Drugs and Crime noted last year that scam centers have been diversifying their worker pools, shifting from predominantly trafficking Chinese nationals and other Chinese speakers to entrapping people from a broader array of countries and backgrounds who speak various languages. UN researchers attributed this change in part to attackers broadening their targets to include different populations around the world. But they added that the dynamic also seemed to be a reaction to Chinese enforcement and Beijing’s efforts to protect Chinese citizens.

“China is doing more to fight fraud—like orders of magnitude more—than any other country,” says Gary Warner, a longtime digital scams researcher and director of intelligence at the cybersecurity firm DarkTower. “But I would agree that the crackdown by China on people scamming China has squeezed the balloon so to speak and led to more international and American targeting.”

The Chinese government has spent years investing in national safety campaigns warning citizens about the threat of scams and how to avoid falling victim to them. Some of the public discourse attempts to appeal to a sense of national solidarity. There’s a common meme in China, 中国人不骗中国人, literally, “Chinese people don’t deceive Chinese people” that is used to signal trust when swapping restaurant recommendations or job leads. In the context of digital scams, a variant has emerged: “Chinese don’t scam Chinese.”

Standing inside the HumanX conference in San Francisco’s Moscone Center, it’s hard not to feel like you’re at the center of the AI universe. Technology leaders swarm the building, and the headquarters of OpenAI and Anthropic are just down the block. But a 70-person startup headquartered 5,000 miles away in Germany’s Black Forest—a region famous for its ham—has become a top competitor to Silicon Valley’s leading labs in AI image generation.

In December, Black Forest Labs raised funds at a $3.25 billion valuation, after signing deals to power AI image-generation features in Adobe and the graphic design platform Canva. It has even struck agreements with major AI labs like Microsoft, Meta, and xAI to power similar features in their products.

Nearly two years after launch, Black Forest Labs can afford to be picky about who it works with. In 2024, Elon Musk’s xAI tapped Black Forest Labs to power Grok’s first image generator. That partnership put Black Forest Labs on the map but generated a lot of controversy due to the chatbot’s limited safeguards. It ended months later when xAI developed an in-house AI image model.

In recent months, xAI approached Black Forest Labs about licensing the startup’s technology again, sources familiar with the matter tell WIRED. This time around, Black Forest Labs declined, the sources said, deeming it too operationally difficult to partner with xAI, which has a famously chaotic work environment. xAI did not immediately respond to WIRED’s request for comment.

In September, Black Forest Labs struck a $140 million multiyear deal to give Meta access to its AI image-generation technology.

These AI labs want to work with Black Forest Labs because its image generators are among the world’s best, ranking just below OpenAI and Google’s offerings on the third-party firm Artificial Analysis’ benchmarks. The startup also offers some of the most downloaded text-to-image models on Hugging Face, indicating that a lot of AI image tools on the market are likely powered by a free version of Black Forest Labs’ technology.

It’s particularly impressive since the company has historically had far fewer resources than its competitors. This has led it to a more efficient line of research called latent diffusion, which is essentially when an AI model first sketches out a rough blueprint of an image, and then paints in more detail.

Latent diffusion “enabled us to put out very powerful models that took orders of magnitude less resources than our competitor’s models,” said cofounder Andreas Blattmann in an interview with WIRED onstage at HumanX this week.

Despite its success, Black Forest Labs believes image generation is just the beginning. Blattmann said the startup plans to unveil a robot powered by one of its AI models later this year. (He did not reveal what company is making the hardware.) The push is part of a larger opportunity the company sees to build AI that can perceive and take actions in the physical world.

“Visual intelligence is so much more than content creation. Content creation is just the first segue into this entire technology,” said Blattmann. “What I’m personally super excited about—and that’s a pattern throughout this conference—is physical AI.”

Black Forest Labs is also in talks with a handful of hardware companies, to power features in products like smart glasses and robots, sources tell WIRED.

Building in the Black Forest

Blattmann and his cofounders, Robin Rombach and Patrick Esser, made a name for themselves publishing some groundbreaking research on AI image models in 2021. In 2022, they were hired by Stability AI and released Stable Diffusion, a popular open source AI image generator based on their prior research. But two years later, they announced their departure and launched Black Forest Labs.

Rather than move to San Francisco, the trio decided to maintain a headquarters near their hometowns in Freiburg, Germany. Blattmann said the decision has been key to the company’s success.

Muscles are remarkably effective systems for generating controlled force, and engineers developing hardware for robots or prosthetics have long struggled to create analogs that can approach their unique combination of strength, rapid response, scalability, and control. But now, researchers at the MIT Media Lab and Politecnico di Bari in Italy have developed artificial muscle fibers that come closer to matching many of these qualities.

Like the fibers that bundle together to form biological muscles, these fibers can be arranged in different configurations to meet the demands of a given task. Unlike conventional robotic actuation systems, they are compliant enough to interface comfortably with the human body and operate silently without motors, external pumps, or other bulky supporting hardware.

The new electrofluidic fiber muscles — electrically driven actuators built in fiber format — are described in a recent paper published in Science Robotics. The work is led by Media Lab PhD candidate Ozgun Kilic Afsar; Vito Cacucciolo, a professor at the Politecnico di Bari; and four co-authors.

The new system brings together two technologies, Afsar explains. One is a fluidically driven artificial muscle known as a thin McKibben actuator, and the other is a miniaturized solid-state pump based on electrohydrodynamics (EHD), which can generate pressure inside a sealed fluid compartment without moving parts or an external fluid supply.

Until now, most fluid-driven soft actuators have relied on external “heavy, bulky, oftentimes noisy hydraulic infrastructure,” Afsar says, “which makes them difficult to integrate into systems where mobility or compact, lightweight design is important.” This has created a fundamental bottleneck in the practical use of fluidic actuators in real-world applications.

The key to breaking through that bottleneck was the use of integrated pumps based on electrohydrodynamic principles. These millimeter-scale, electrically driven pumps generate pressure and flow by injecting charge into a dielectric fluid, creating ions that drag the fluid along with them. Weighing just a few grams each and not much thicker than a toothpick, they can be fabricated continuously and scaled easily. “We integrated these fiber pumps into a closed fluidic circuit with the thin McKibben actuators,” Afsar says, noting that this was not a simple task given the different dynamics of the two components.

A key design strategy was to pair these fibers in what are known as antagonistic configurations. Cacucciolo explains that this is where “one muscle contracts while another elongates,” as when you bend your arm and your biceps contract while your triceps stretch. In their system, a millimeter-scale fiber pump sits between two similarly scaled McKibben actuators, driving fluid into one actuator to contract it while simultaneously relaxing the other.

“This is very much reminiscent of how biological muscles are configured and organized,” Afsar says. “We didn’t choose this configuration simply for the sake of biomimicry, but because we needed a way to store the fluid within the muscle design.” The need for an external reservoir open to the atmosphere has been one of the main factors limiting the practical use of EHD pumps in robotic systems outside the lab. By pairing two McKibben fibers in line, with a fiber pump between them to form a closed circuit, the team eliminated that need entirely.

Another key finding was that the muscle fibers needed to be pre-pressurized, rather than simply filled. “There is a minimum internal system pressure that the system can tolerate,” Afsar says, “below which the pump can degrade or temporarily stop working.” This happens because of cavitation, in which vapor bubbles form when the pressure at the pump inlet drops below the vapor pressure of the liquid, eventually leading to dielectric breakdown.

To prevent cavitation, they applied a “bias” pressure from the outset so that the pressure at the fiber pump inlet never falls below the liquid’s vapor pressure. The magnitude of this bias pressure can be adjusted depending on the application. “To achieve the maximum contraction the muscle can generate, we found there is a specific bias pressure range that is optimal,” she says. “If you want to configure the system for faster response, you might increase that bias pressure, though with some reduction in maximum contraction.”

Cacucciolo adds that most of today’s robotic limbs and hands are built around electric servo motors, whose configuration differs fundamentally from that of natural muscles. Servo motors generate rotational motion on a shaft that must be converted into linear movement, whereas muscle fibers naturally contract and extend linearly, as do these electrofluidic fibers. 

“Most robotic arms and humanoid robots are designed around the servo motors that drive them,” he says. “That creates integration constraints, because servo motors are hard to package densely and tend to concentrate mass near the joints they drive. By contrast, artificial muscles in fiber form can be packed tightly inside a robot or exoskeleton and distributed throughout the structure, rather than concentrated near a joint.”

These electrofluidic muscles may be especially useful for wearable applications, such as exoskeletons that help a person lift heavier loads or assistive devices that restore or augment dexterity. But the underlying principles could also apply more broadly. “Our findings extend to fluid-driven robotic systems in general,” Cacucciolo says. “Wherever fluidic actuators are used, or where engineers want to replace external pumps with internal ones, these design principles could apply across a wide range of fluid-driven robotic systems.”

This work “presents a major advancement in fiber-format soft actuation,” which “addresses several long-standing hurdles in the field, particularly regarding portability and power density,” says Herbert Shea, a professor in the Soft Transducers Laboratory at Ecole Polytechnique Federale de Lausanne in Switzerland, who was not associated with this research. “The lack of moving parts in the pump makes these muscles silent, a major advantage for prosthetic devices and assistive clothing,” he says.

Shea adds that “this high-quality and rigorous work bridges the gap between fundamental fluid dynamics and practical robotic applications. The authors provide a complete system-level solution — characterizing the individual components, developing a predictive physical model, and validating it through a range of demonstrators.”

In addition to Afsar and Cacucciolo, the team also included Gabriele Pupillo and Gennaro Vitucci at Politecnico di Bari and Wedyan Babatain and Professor Hiroshi Ishii at the MIT Media Lab. The work was supported by the European Research Council and the Media Lab’s multi-sponsored consortium.