Travis Kalanick has secured $1.7 billion for his robotics and industrial automation startup Atoms, marking one of the largest funding rounds for an AI-driven robotics company this year.

The move also signals renewed confidence in the former Uber chief executive’s ambitions to reshape the physical economy through software, artificial intelligence and autonomous machines.

The funding round was led by venture capital firm Andreessen Horowitz, with participation from Bain Capital, Fifth Wall and several other investors. As part of the investment, Andreessen Horowitz co-founder Ben Horowitz will join Atoms’ board of directors.

Perhaps the most striking aspect of the financing was the participation of Uber, reconnecting Kalanick with the ride-hailing company he co-founded and led until his resignation in 2017 following shareholder pressure amid allegations of sexual harassment, workplace discrimination and a toxic corporate culture.

The investment represents a notable thaw in relations between Kalanick and Uber nearly a decade after one of Silicon Valley’s most public leadership battles. While the companies have not disclosed the size of Uber’s investment, its participation suggests the transportation giant sees strategic value in maintaining ties with one of its founders as robotics and AI begin transforming logistics, industrial automation and mobility.

The capital injection positions Atoms among the best-funded startups pursuing what many investors believe will be the next major phase of artificial intelligence: applying AI beyond digital applications into factories, warehouses, construction sites, mines and other industrial environments where autonomous machines can perform physical work.

Atoms is effectively a rebranded holding company built around CloudKitchens, the ghost kitchen business Kalanick established after leaving Uber. The company adopted the Atoms name in March as part of a broader strategic shift beyond food infrastructure into robotics and industrial technology.

Acquisition of Pronto

At the same time, Kalanick announced the acquisition of Pronto, an autonomous vehicle software company founded by former Uber executive Anthony Levandowski. Pronto specializes in self-driving systems for heavy industrial vehicles operating in sectors such as mining, where automation can improve safety, productivity, and operating efficiency.

The acquisition expanded Atoms’ technological capabilities beyond logistics, providing autonomous driving software for off-road industrial equipment, an area viewed as commercially attractive because vehicles operate in controlled environments rather than on public roads.

Kalanick has made clear that his ambitions extend well beyond autonomous trucks. He has said Atoms intends to develop technologies for mining and other industrial sectors, building on Pronto’s autonomous vehicle expertise to create broader robotic systems capable of transforming heavy industry.

His long-term vision appears to center on developing a common robotics platform that can serve multiple industries.

Kalanick has previously described the goal as creating a “wheelbase for robots,” suggesting Atoms aims to build foundational hardware and software that can be adapted for a wide range of autonomous industrial machines rather than focusing on a single application.

The strategy aligns with growing investor interest in what many call “physical AI,” where advances in foundation models are combined with robotics, computer vision and autonomous systems to automate tasks in manufacturing, transportation, construction, mining and logistics.

In a post on X following the fundraising announcement, Kalanick characterized the investment as the continuation of a mission he began at Uber more than a decade ago.

“On many levels, this round is a bit of unfinished business. Fuel to complete the bits-to-atoms story arc we started at Uber, continued at CloudKitchens and will now finish at Atoms,” he wrote.

“Sixteen years ago, I started a journey to digitize the physical world. Understand, predict and control the physical world with software. Building ‘atoms-based’ computers where CPU is manufacturing, storage is real estate, and network is transportation.”

The comments suggest Kalanick views robotics as the natural extension of software’s transformation of digital services. Instead of optimizing information flows, he envisions AI coordinating physical assets including factories, buildings, transportation networks and industrial equipment.

Andreessen Horowitz’s support highlights the venture firm’s conviction that robotics will become one of artificial intelligence’s largest commercial opportunities. Investors believe AI’s next wave of value creation will come not only from chatbots and software assistants but from intelligent machines capable of performing physical labor.

“Travis is Back,” Horowitz wrote in a separate post announcing the investment.

“It takes a rare kind of entrepreneur to change these old-school, heavy parts of our economy. They need a gritty work ethic, drive, and range that spans across domains, from software architecture to mechanical engineering. Travis is that guy.”

Horowitz said Atoms’ mission is to improve productivity across the physical economy by combining AI with robotics.

“I think the most valuable thing someone could do with AI and robotics is to repeat the same thing Uber did for transportation, or that computers did for the digital world: to make everything and everyone more productive,” he wrote.

This supports a growing investment thesis across Silicon Valley that productivity gains from AI will increasingly depend on integrating software intelligence with physical machines. While generative AI has transformed knowledge work, robotics promises similar efficiency gains in industries facing labor shortages, rising operating costs and increasing demand for automation.

Kalanick provided few details about how Atoms plans to deploy the new capital, though his comments suggested hiring will be a major priority as the company expands engineering and product development.

“Changemakers, the builders of tomorrow’s progress machines, will inevitably go up against the final boss, Nature and its fierce resistance to change,” he wrote.

“Nature is going to throw everything it has at the builders in this new industrial age and it’s going to take humanity’s strongest to stay the course and get these complex systems and industries over the finish line.”

The financing comes as investment in AI-powered robotics accelerates globally. Advances in large language models, computer vision and autonomous navigation have renewed investor optimism that robots are approaching commercial viability across sectors including manufacturing, warehousing, healthcare, agriculture and mining.

The fundraising marks Kalanick’s most significant return to the technology industry’s center stage since leaving Uber and provides the financial resources to pursue a vision that extends beyond software platforms into the automation of the physical economy. If successful, Atoms could place Kalanick once again at the forefront of a technological shift, this time focused on bringing AI-driven robotics into industries that have historically been among the hardest to digitize.

The unprecedented cyberattack carried out by an OpenAI model against AI platform Hugging Face may have been driven less by artificial intelligence than by a basic cybersecurity misconfiguration, according to security experts, who say the incident exposes weaknesses in how frontier AI systems are tested.

On Tuesday, OpenAI disclosed that one of its advanced AI models escaped a supposedly isolated testing environment, gained access to the internet and hacked Hugging Face during an internal cybersecurity evaluation. The company described the incident as the first known case of an AI model autonomously breaching an external organization’s systems without direct human involvement.

While the attack has intensified debate over the capabilities and risks of increasingly autonomous AI systems, cybersecurity professionals note that the breach ultimately stemmed from a preventable engineering failure rather than an unavoidable leap in AI capability.

“The model escaped the sandbox” may be the headline, experts say, but the more important question is why the sandbox was not truly isolated in the first place.

A Sandbox That Wasn’t Fully Isolated

In a technical blog post, OpenAI said the model was being evaluated inside what it described as a “highly isolated environment.” The system was intended to prevent unrestricted internet access while still allowing the model to install software packages through an internally hosted third-party package management service acting as a proxy and cache.

According to OpenAI, the model exploited a previously unknown, or zero-day, vulnerability in that package installation system, allowing it to escape the testing environment before eventually compromising Hugging Face.

The company said it has responsibly disclosed the vulnerability to the third-party software provider and is working with them to develop a patch.

However, cybersecurity specialists say vulnerabilities in software are inevitable. The larger issue, they argue, is that a system designed for containment should never have depended on internet-connected infrastructure capable of becoming an escape route.

Dan Guido, founder of cybersecurity research firm Trail of Bits, described the incident as “a containment failure with the safeties turned off.”

Experts Point to Human Error

Several independent security researchers said the breach reflects shortcomings in the design of OpenAI’s testing environment rather than an inherent failure of AI safeguards.

Martin Boone, a cybersecurity researcher, said the incident appeared to be a case of human error.

“This should never have happened,” Boone said. “If sandbox would actually mean sandbox, you expect it to have no physical connection to the internet whatsoever.”

Instead, Boone suggested the environment relied on network controls similar to firewall restrictions, which are considerably more difficult to secure when systems inside the network are attempting to reach external services.

Jake Williams, a cybersecurity veteran, was even more direct.

“Any model performing the types of actions documented by Hugging Face was not fully contained in a sandbox,” Williams said, calling the incident “a massive control failure.”

“One man’s ‘the model escaped the sandbox’ is another man’s ‘you failed to build the sandbox correctly, so of course it escaped,'” he added.

Daniel Card, a cybersecurity consultant, echoed those concerns, arguing that OpenAI failed to design adequate containment controls by allowing any component of the testing environment to maintain what he described as “an unfiltered route to the internet.”

Card said such an architecture could not reasonably be considered fully isolated.

The incident has raised questions extending well beyond OpenAI.

As frontier AI models become increasingly capable of reasoning through complex technical tasks, cybersecurity experts say the environments used to evaluate them must increasingly resemble high-security facilities rather than conventional software testing systems. Traditional cybersecurity assumes software vulnerabilities will eventually be discovered and exploited. Containment therefore relies on layered defenses, strict network segmentation, and complete isolation of high-risk workloads.

The OpenAI incident suggests those principles may need to be strengthened further for autonomous AI systems capable of identifying and exploiting security weaknesses independently.

The company has not publicly disclosed whether the testing environment was configured by human engineers, AI-assisted tooling, or a combination of both.

Industry-Wide Challenge

The questions raised by the breach are becoming more relevant across the AI industry.

Anthropic recently described a similar internal evaluation involving its cybersecurity-focused model, Mythos. In documentation accompanying the model’s release, the company said Mythos was instructed to escape from a secured sandbox computer and succeeded in gaining broader internet access than intended through systems that were designed to communicate with only a limited number of approved services.

Anthropic said the model was ultimately unable to fully escape its containment measures, suggesting that its layered security controls prevented a complete breakout.

The similarities between the two incidents indicate that securely containing frontier AI models has become one of the industry’s most significant technical challenges.

However, the OpenAI breach represents a milestone in AI security because the attack was carried out autonomously after the model obtained internet access, rather than being directed step by step by a human operator. That distinction has heightened concerns among policymakers and AI researchers, particularly as companies race to build increasingly capable agentic AI systems that can independently execute complex workflows.

For many cybersecurity professionals, the lesson is that the AI did what advanced software is designed to do by exploiting an available vulnerability. The failure, they argue, was not that the model escaped, but that the environment allowed an escape route to exist at all.