Listing consumer electronics on the internet’s large ecommerce marketplaces is a key step in “democratizing” the products, allowing them to be purchased by anyone with just a click. It has happened to cars (in the United States, you can buy a Hyundai on Amazon), and now it’s happening to humanoid robots.

The Chinese manufacturer Unitree Robotics, among the most active robot-makers in the field, is preparing to bring its most affordable model, the Unitree R1, to international markets through Alibaba Group’s marketplace. According to reports in The South China Morning Post, the rollout will initially cover North America, Japan, Singapore, and Europe. There’s no exact on-sale date for the robots yet, but the Post report says it will show up as soon as this week.

This is not the first time Unitree has used AliExpress as a global storefront. The company’s G1 model, the more powerful and more expensive predecessor to the R1, is already listed at just under $19,000.

The G1 is already on sale on AliExpress.

Advertisement

It’s as much of a symbolic step before as a commercial one; selling a humanoid robot on a global marketplace positions the product as easily attainable. This serves as a step toward normalization of the tech, which is still not widely adopted. The sale of the R1 simply lowers the threshold of access even further, and shifts humanoid robots from the territory of promise to that of concrete availability.

Lower Price, Higher Demand

When it was announced last summer, the starting price of the R1 was 39,900 yuan, or about $5,900. Today, the basic version starts at 29,900 yuan, or about $4,370.

That price will fluctuate given changes in exchange rates and shipping costs that add on import taxes and tariffs. Still, that figure sounds surprisingly low considering that some of the R1’s other competitors in the humanoid robotics landscape are far more expensive.

The price tag for Unitree’s own flagship H1 robot approaches $90,000. Tesla’s Optimus robot, which is not yet on sale to the public, is aiming for a starting price under $20,000, but that price will only be attainable when Tesla reaches production of 1 million units a year. Meanwhile, robots from Figure AI and Apptronik are hovering around $50,000 per unit. The R1’s objectively low price essentially makes it a hatchback in a world of sedans.

The R1 is 4 feet tall, weighs 50 pounds, and has 26 smart joints. You can talk to it and give it commands; Unitree’s large-language multimodal model with voice and image recognition is on board. Curious coders can program it using a software developer’s kit. But the real calling card is the R1’s physical performance. The robot can do cartwheels, lie down and stand up independently, and run downhill. Unitree calls it “born for sport,” and videos of its presentation made the rounds months ago. Handstands and wheel kicks are not exactly what you’d expect from a robot that costs less than a used car.

Put It to Work

As impressive as the Unitree R1’s moves are, it lacks hands with articulated fingers, and its motors can’t generate a lot of torque. It is not designed to be a domestic helper or to manipulate complex objects. The company presents it as an “intelligent companion” for interaction, research, and software development.

The EDU model (Go2 EDU, G1 EDU) add an Nvidia Jetson Orin module with more computing power for artificial intelligence tasks. That model also has two degrees of freedom for the head and optional right hands. In that robot’s case, the target market is laboratories and universities. The limitations of the basic R1 put it largely in the same camp. This is not a household robot that makes coffee and walks the dog, but it is a good choice for researchers, labs, and anyone who wants to test robotics algorithms on solid hardware without spending a fortune.

It is true that bringing a relatively capable humanoid to global markets at this price does lower the barrier to entry for developers, researchers, and enthusiasts. It is a real leap from a few years ago, even if some people will buy it just to keep it in the living room to take a bow when guests arrive.

This story was originally published by WIRED Italia and translated from Italian.

Long ago, in the aftermath of the UNIX wars, three kernels emerged from the rubble: BSD, Linux, and Hurd. BSD, being UNIX, was held back by legal wrangling in the aftermath of the wars, and that allowed Linux to pull ahead to a pole position it still enjoys to this day. BSD has its following, of course, but Hurd? GNU Hurd seemed destined to languish… until April 1st, 2026, when the Gentoo Linux distribution was ported to the Free Software Foundation’s kernel.

It turns out, they weren’t actually joking. The joke part was that they were moving fully to the Hurd kernel, away from Linux– you can absolutely still run Gentoo with the Linux kernel, and make no mistake, that’s still the default and best-supported option. Options are good, though, and the Gentoo team has decided that it’s time to add some options to the kernel space, and give the Hurd some time in the sun.

Unlike the Linux kernel, which follows closely the monolithic UNIX framework– and the BSD-Unix kernel, which is Unix–GNU Hurd is a microkernel architecture, based originally on the Mach kernel. In that, it’s rather like MacOS. Unlike MacOS, given its roots in the Free Software Foundation, GNU Hurd is 100% free and open source. There are advantages to a microkernel architecture– it keeps drivers out of kernel space so a dodgy WiFi adapter can’t crash your system, for example– but the big disadvantage is, of course, drivers. Both Linux and BSD drivers can be ported, but that takes work and many of them have not been.

Still, now that Microsoft has become a major contributor to the Linux kernel, we could see a lot of the old-school Linux users who talk about “win-doze” and still spell Microsoft with a dollar sign being tempted to join the Hurd. If that appeals to you and you’re not into Gentoo, Debian has quietly let you install with the Hurd kernel for years now. It’s either that or embrace BSD and escape the chaos vortex.

The big three aren’t the only POSIX kernels out there, of course– there’s even one written entirely in Rust, for the die hard rustaceans amongst you.

Over the past couple of years, Google has found ways to stuff Gemini in nearly every app and service it offers. Whether it’s Gmail with its AI inbox or Chrome with its chat sidebar, Gemini is now inescapable inside of Workspace. I don’t know about you, but I don’t need an AI to tell me how to write a =SUM equation in Sheets or an outline for a first draft. Most of the time, I find Gemini is a distraction. If you feel the same way, this how-to is for you.

How to remove Gemini from your personal Workspace account

To turn Gemini off, you will need to disable two separate sets of options. The first set covers a set of features, including smart compose, that are shared across Gmail, Chat and Meet — so if you turn them off in one app, they won’t be available in any of the three. All of this is most easily done through Gmail’s web client.

1. In Gmail, tap the cog icon.

3. Under the General tab, scroll down to find Smart features.

4. Disable Turn on smart features in Gmail, Chat, and Meet.

In Japan, Switzerland, the United Kingdom or European Economic Area, smart features are turned off by default.

Next, turn your attention to Workspace.

1. In Gmail, tap the cog icon.

3. Under the General tab, scroll down and click Manage Workspace smart feature settings.

4. Toggle off smart features in Google Workspace and Smart features in other Google products.

A word of warning: completely disabling smart features in Google Workspace turns off not only Gemini integration but also access to basic capabilities like spelling and grammar corrections. You’ll also lose features that have been Google staples for years. In Gmail, for example, the app will stop sorting incoming emails by priority, a notification at the top of the screen informing you that smart features are required for inbox categorization. Whatever Google’s motivation for this state of affairs, it’s a design decision that actively discourages users from disabling Gemini integration.

If you want to rid yourself of Gemini but would still like to use some of the other features the company offers through Gmail and its other apps, I recommend leaving the first set of smart features on while disabling the Workspace-specific ones. You can also opt to turn off some of the features included in the first group, while leaving others on. Below is a list of those features, with a brief overview of the less self-explanatory ones.

• Smart Compose — as you write an email, Gmail will generate predictive writing suggestions

• Smart Compose personalization — as you write, Gmail will tailor Smart Compose suggestions to your writing style

• Nudges — Gmail will generate notifications prompting you to respond or follow up on unanswered emails

• Smart Reply — Gmail will generate suggestions on how to respond to an email

• Package tracking — Google will display shipping updates inside of Gmail

• Desktop notifications — Yes, for some reason you need the power of AI to get notifications on your PC

Unfortunately, Google doesn’t offer this same level of granular control when it comes to smart features inside of Workspace. For instance, if you turn off Gemini in Docs, Calendar won’t automatically display events from Gmail. Again, Google really wants to dissuade you from disabling Gemini.

How to remove Gemini from your professional Workspace account

If your workplace uses Google Workspace, all of the above options should be present in Gmail’s settings menu, and you can follow the same steps to turn off most of the smart features Google offers. Unfortunately, the second part of the process does nothing. You will still see Gemini in Docs, Sheets and elsewhere, even with smart features in Workspace turned off. Only your admin can completely turn off Gemini for you.

Presented by Edgeverve

Smart, semi‑autonomous AI agents handling complex, real‑time business work is a compelling vision. But moving from impressive pilots to production‑grade impact requires more than clever prompts or proof‑of‑concept demos. It takes clear goals, data‑driven workflows, and an enterprise platform that balances autonomy, governance, observability, and flexibility with hard guardrails from day one.

From pilots to the “operational grey zones”

The next wave of value sits in the connective tissue between applications — those operational grey zones where handoffs, reconciliations, approvals, and data lookups still rely on humans. Assigning agents to these paths means collapsing system boundaries, applying intelligence to context, and re‑imagining processes that were never formally automated. Many pilots stall because they start as lab experiments rather than outcome‑anchored designs tied to production systems, controls, and KPIs.

Start with outcomes, not algorithms. Translate organizational KPIs (cash‑flow, DSO, SLA adherence, compliance hit rates, MTTR, NPS, claims leakage, etc.) into agent goals, then cascade them into single‑agent and multi‑agent objectives. Only after goals are explicit should you select workflows and decompose tasks.

Pick targets, then decompose the work

What does “target” actually mean? In agentic programs, a target is a business outcome and the use case that moves it. For example, “reduce unapplied cash by 20%” target outcome; “cash application and exceptions handling” use case. With the use case in hand, perform persona‑level task decomposition: map the human role (e.g., cash applications analyst, facilities coordinator), enumerate their tasks, and identify which are ripe for agentification (data retrieval, matching, policy checks, decision proposals, transaction initiation).

Delivering on those tasks requires a data‑embedded workflow fabric that can read, write, and reason across enterprise systems while honoring permissions. Data must be AI‑ready, discoverable, governed, labeled where needed, augmented for retrieval (RAG), and policy‑protected for PII, PCI, and regulatory constraints.

Integration goes beyond APIs

APIs are one mode of integration, not the only one. Robust agent execution typically blends:

• Stable APIs

  with lifecycle management for core systems

  Advertisement

• Event‑driven triggers

  (streams, webhooks, CDC) to react in real time

• UI/RPA fallbacks

  where APIs don’t exist

• Search/RAG connectors

  for documents and knowledge bases

• Policy management

  across tools and actions to enforce entitlements and segregation of duties

  Advertisement

The north star is integration reliability — built on idempotency, retries, circuit-breakers, and standardized tool schemas — so agents don’t “hallucinate” actions the enterprise can’t verify.

A quick example: finance and facilities, in production

Inside our organization, we deployed specialized agents in a live CFO environment and in building maintenance. In finance, seven agents interacted with production systems and real accountability structures. Year‑one outcomes included: >3% monthly cash‑flow improvement, 50% productivity gain in affected workflows, 90% faster onboarding, a shift from account‑level handling to function‑level orchestration, and a $32M cash‑flow lift. These results don’t guarantee gains everywhere; they show that designing products can deliver measurable outcomes on a scale.

The four design pillars: Autonomy, governance, observability & evals, flexibility

1) Autonomy: right‑size it to the risk

Autonomy exists on a spectrum. Early efforts often automate well‑bounded tasks; others pursue research/analysis agents; increasingly, teams target mission‑critical transactional agents (payments, vendor onboarding, pricing changes). The rule: match autonomy to risk, and encode the operating mode suggest‑only, propose‑and‑approve, or execute‑with‑rollback per task.

2) Governance: guardrails by design, not as bolt‑ons

Unbounded agents create unacceptable risk. Build guardrails into the plan:

• Policy & permissions: tie tools/actions to identity, scopes, and SoD rules.

• Human‑in‑the‑loop (HITL): where mission‑critical thresholds are crossed (amount, vendor risk, regulatory exposure).

• Agent lifecycle management: versioning, change control, regression gates, approval workflows, and sunsetting.

• Third‑party agent orchestration: vet external agents like vendors, capabilities, scopes, logs, SLAs.

• Incident and rollback: kill‑switches, safe‑mode, and compensating transactions. This is how you

  scale innovation safely while protecting brand, compliance, and customers.

3) Observability & evaluations: trust comes from telemetry

Production agents need the same rigor as any core platform:

• Telemetry: capture full execution traces across perception, planning, tool use, action supported by structured logs and replay.

• Offline evals: cenario tests, red‑teaming, bias and safety checks, cost/performance benchmarks; baseline vs. challenger comparisons.

• Online evals: shadow mode, A/B, canary releases, guardrail breach alerts, human feedback loops.

• Explainability & auditability: why was an action taken, which data/tools were used, and who approved.

4) Flexibility: assume volatility, design for swap‑ability

Models, tools, and vendors change fast. Treat agentic capability as platform currency: create an environment where teams can evaluate, select, and swap models/tools without tearing down the build. Use a model router, tool registry, and contract‑first interfaces so upgrades are controlled experiments, not rewrites.

The agent platform fabric: how platformization turns goals into outcomes

A true agentic enterprise requires a platform fabric that transforms goals into outcomes, not a patchwork of isolated pilots. This platform anchors enterprise‑to‑agent KPI cascades, drives task decomposition and multi‑agent planning, and provides governed tooling and data access across APIs, RPA, search, and databases.

It centralizes knowledge and memory through RAG and vector stores, enforces enterprise controls via a policy engine, and manages performance and safety through a unified model layer. It supports robust orchestration of first‑ and third‑party agents with common context, embeds deep observability and evaluation pipelines, and applies disciplined release engineering from sandbox to GA. Finally, it ensures long‑term resilience through lifecycle management versioning, deprecation, incident playbooks, and auditable histories.

Guardrails in action: a BFSI example

Consider payments exception handling in banking — high stakes, regulated, and customer‑visible. An agent proposes a resolution (e.g., auto‑reconcile or escalate) only when:

• The transaction falls below risk thresholds; above them, it triggers HITL approval.

• All policy checks (KYC/AML, velocity, sanctions) pass.

• Observability hooks record rationale, tools invoked, and data used.

• Rollback/compensation is defined if downstream failures occur. This pattern generalizes to vendor onboarding, pricing overrides, or claims adjudication — mission‑critical work with explicit safety rails.

Scale beyond pilots

Scaling agentic AI beyond pilots demands disciplined readiness across nine fronts: leaders must clarify which KPIs matter and how agent goals ladder into them, determine which persona tasks are agentified versus remain human‑led, and align each with the right autonomy mode from suggest‑only to propose‑and‑approve to execute‑with‑rollback. They must embed governance guardrails, including HITL points and lifecycle controls; ensure robust observability and evaluation via telemetry, replay, audits, and offline/online tests; and verify data readiness, with governed, policy‑protected, retrieval‑augmented data flows. Integration must be reliable, with API lifecycle management, event triggers, and RPA/other fallbacks. The underlying platform should enable model swap‑ability and orchestration of first‑ and third‑party agents without rebuilding. Finally, measurement must focus on true operational impact cash flow, cycle times, quality, and risk reduction rather than task counts.

The takeaway

Agentic AI is not a shortcut; it’s a new system of work. Enterprises that approach it with platform discipline aligning autonomy with risk, embedding governance and observability, and designing for swap‑ability will convert pilots into production impact. Those that don’t keep accumulating impressive but disconnected demos. The difference isn’t how fast you ship an agent; it’s how deliberately you design the enterprise around it.

N. Shashidar is SVP & Global Head, Product Management at EdgeVerve.

Sponsored articles are content produced by a company that is either paying for the post or has a business relationship with VentureBeat, and they’re always clearly marked. For more information, contact sales@venturebeat.com.

Started by three NUS students, Juno Jane gained traction through TikTok virality

Not many students can run a business alongside their studies, but three Gen Z founders have found a way to make it work.

Sophia Poh, 24, Kayleigh Low, 22, and Wanzhen Li, 22, are the trio behind Juno Jane, a Singapore-based handbag brand they founded while studying at the National University of Singapore (NUS).

The idea was born from classroom conversations, which eventually grew into a full-fledged business that has since achieved six-figure revenue and shipped to over 40 countries, all while the founders balanced university coursework.

The market is saturated, but they still found an opening

Sophia, Kayleigh, and Wanzhen met in a business class at NUS in early 2024. A casual conversation about the lack of durable, aesthetic, and functional handbags for young professional women sparked the idea that would eventually become their brand.

Despite the saturated market, they spotted an opening: many brands were going for speed and prioritising the launch of more products, instead of designing features with the customer in mind.

Conversely, the Juno Jane founders wanted to take a different approach: one that was design-led and focused on tailoring their bags to what consumers actually wanted.

However, without any prior experience, the process was challenging. They surveyed friends, interviewed peers to identify common pain points, and scoured the Internet for manufacturer contacts. Eventually, they connected with one, working closely with the manufacturer to refine sketches and bring their ideas to life.

The trio saved up their internship allowance to launch Juno Jane

As students, the founders pooled together five figures from their internship savings to launch Juno Jane in 2024. While the bags were designed in-house by the founders, they are produced at a factory in China, which they sourced through a combination of personal connections and online research.

Most of the trio’s capital went towards research and development, as well as producing samples and prototypes—a process Wanzhen noted “is much more expensive than the actual final cost of the bag because of all the behind-the-scenes testing.”

In fact, the trio shared that their first collection, the Kai Tote, went through six iterations over the course of a year before launch.

Their meticulous approach paid off, though.

The Kai Tote sold out within two weeks of its Dec 2024 launch. And shortly after, Juno Jane broke even in Feb 2025 following its first restock. The collection was subsequently expanded in Apr 2025 with black and wine colourways.

Gaining traction through TikTok

A big part of this early success came from organic community building. From the start, the founders documented their journey on TikTok, sharing behind-the-scenes content of factory visits, design decisions, and the realities of student entrepreneurship.

“When we started, we already launched to a community who knew who we were,” Kayleigh said. “That’s really what helped to sell out our first collection.”

To date, Juno Jane’s videos have garnered over 340,000 likes on TikTok alone, alongside a following of over 40,000 on both Instagram and TikTok.

Their most viral video has garnered over 33,000 likes, though not all their posts achieved traction, especially in the early days. Instead, the founders gradually built momentum by sharing candid stories as young entrepreneurs.

Juno Jane’s videos eventually began reaching international audiences, and today, overseas sales comprise 40% of its revenue, primarily from the United States, Australia, and the Philippines.

Running a business is nothing like studying it in class

While three founders studied business at NUS, they quickly discovered that running a company is different from theories taught in school.

Instead, running a business requires integrating cross-functional skills that university courses teach in isolation. 

“In school, when you learn about pricing, it’s just pricing alone,” Kayleigh observed. “But in real business, you see how supply chain affects marketing—if our bags are on pre-order, our marketing effectiveness drops because people cannot buy immediately.”

That’s why the team now orders stock in advance and operates two warehouse units to store inventory, ensuring products are readily available for purchase, and customers don’t have to wait.

Priced in the mid-hundreds, Juno Jane’s bags are designed to appeal to a wide range of customers. “We didn’t want to design bags that could only be sold to people willing to pay S$300 or S$400,” Kayleigh added, noting that they aimed to strike a balance between accessibility, functionality, and style.

Competing against established players

Since launch, Juno Jane has generated over six figures in gross sales. Against established players, it competes by staying lean and design-focused.

The trio split responsibilities across complementary skill sets. Kayleigh handles the Juno Jane website and TikTok, Wanzhen manages Instagram and creative design, while Sophia oversees finance and operations. This structure allows them to cover all business functions without hiring, managing expenditure as a bootstrapped business.

Moreover, their low customer acquisition costs, driven by strong organic TikTok content, provide a key advantage over well-funded competitors.

“We’re very heavy on organic marketing,” Sophia noted. “Because of that, the cost per acquisition for us is still relatively low even though the spend is lower.”

However, like many growing direct-to-consumer brands, Juno Jane initially struggled with demand forecasting. In the beginning, they were sometimes out of stock for two months due to production periods. They have since implemented a system to track sales weekly and place orders in advance, factoring in manufacturers’ peak periods, like the Chinese New Year break.

“It’s still something we’re learning,” Kayleigh admitted, “because when you’re trying to grow a business, your turnover rate is not linear.”

While e-commerce remains their primary channel, the trio recognise the value of physical touchpoints that help to grow brand awareness and learn from other founders.

Beyond selling products at Gathery and NUS Business School’s Homecoming, they led an initiative called For Her once in the past year at Kada. It is a pop-up event focused on female-led businesses and growing a community with local brands.

“We decided to bring these brands together to create a touch point for everyone to get to know each other better,” Sophia explained. “Nowadays, when we have problems regarding the operational side of things or shipping, we just reach out to each other.”

A retail store is not on the cards just yet

Although the brand has achieved significant growth, the founders shared that opening a permanent retail store is not on the cards just yet, citing the high overhead costs involved. Additionally, two of the founders are still studying at NUS, while Sophia, who graduated last year, is now working full-time.

Instead, they hope to expand their physical presence through pop-ups and consignment arrangements. They are also developing accessories to better cater to their customers’ evolving preferences.

Along the way, the trio learned several lessons over the past two years. “Taking the risk isn’t as scary as you think it might be, especially if you have a team to do it with,” Sophia reflected.

“The inertia is real—like, I have to pull X amount of money and what if it doesn’t work out? But the truth is you don’t really need such a big pool to start with, and with two to three people, you’re kind of de-risking as well.”

Juno Jane’s journey shows that starting a business doesn’t require perfect timing or large capital—just a willingness to take calculated risks with the right team. By sharing responsibilities, staying lean, and testing ideas directly with customers, the founders were able to de-risk entrepreneurship while still moving quickly.

• Learn more about Juno Jane here.
• Read other articles we’ve written on Singaporean businesses here.

OpenAI has acquired personal finance startup Hiro Finance, founder Ethan Bloch announced on Monday and OpenAI confirmed to TechCrunch. The startup was backed by A-list fintech VC firm Ribbit, as well as General Catalyst and Restive.

Terms of the acquisition were not disclosed, nor did Hiro ever disclose how much money it raised. Since Hiro said it will be shutting down its operations on April 20 and deleting all data from its servers on May 13, we’re going to call this an acquihire.

Bloch said in his post that Hiro employees are coming with him to OpenAI. He didn’t specify how many employees that entails, but LinkedIn lists about 10 people associated with the company. Bloch did not respond to our request for comment.

The company was founded in 2023 and launched its AI tool about five months ago. Hiro offered AI-powered financial planning for consumers. Users entered financial information like salary, debts, and monthly costs, and the app modeled different what-if scenarios to help them make financial decisions.

Hiro was specifically trained to nail financial math, including an option that allowed users to verify accuracy, Bloch said in a demo of the product. Over the past couple of years, state-of-the-art frontier models have gotten significantly better (even good) at math of all kinds. But historically, they haven’t been.

This deal stands out for a couple of reasons. Bloch previously founded Digit, a digital-only bank that helped people automatically save money. Digit was sold to Oportun in 2021 for more than $200 million, according to Oportun.

Plus, this isn’t the first financial app OpenAI has bought. Given that OpenAI markets ChatGPT as a good tool for business finance teams, we can see why the model maker would be looking to add more talent to this side of the house. Whether OpenAI plans to pursue financial planning as a more specialized app, we’ll have to wait and see.

It’s also possible that this acquihire is an effort to make OpenAI more popular with OpenClaw users, who often tend to prefer Claude. OpenClaw is a popular agent for robo stock trading. In fact, Bloch created his own auto-trading OpenClaw agent that he named RoboBuffett, he said on LinkedIn.

Another fun fact: Bloch told Business Insider that Hiro was the 15th project he launched, having started as a tech entrepreneur when he was a 13-year-old. The first 13 failed, he said. He sold No. 14, Flowtown, a social media SaaS tool launched in 2009, for $4.5 million. Bloch said he sold Digit for about $230 million. Now he’s sold his latest startup to OpenAI, a company that has broken records for growth and raising money, and may yet break records with an IPO.

An Amazon employee at the Troutdale, Oregon warehouse passed away at work last week, a company spokesperson confirmed to TechCrunch.

According to a report from the Western Edge, an independent investigative outlet covering the Pacific Northwest, the worker collapsed on the floor at the PDX9 warehouse and lay dead as employees continued to work around him.

“We’re deeply saddened by the passing of a member of our team, and our thoughts and deepest sympathies are with their loved ones during this difficult time,” Amazon spokesperson Sam Stephenson told TechCrunch. “We’ve been in touch with his family and have provided resources to support them. For employees at our PDX9 facility, we’ve provided onsite grief counselors and additional support. We’re thankful for the work of the Multnomah County Sherrif’s Department and local emergency medical services.”

On a Reddit forum for Amazon fulfillment center workers, several people claiming to work at PDX9 said that the building had been especially hot after soundproof curtains were installed, which limited airflow. They speculated that the heat could have contributed to the employee’s death, as it would compound the physical demands of fulfillment center work. According to the Western Edge, some employees noticed that the building was cooler when they returned to work the next day.

Amazon, however, said that Oregon’s Occupational Safety and Health Administration (OSHA) determined the incident to be non-work related. Employees were sent home early and were paid for the remainder of their shift; the night shift was cancelled, and employees scheduled to work were paid as well, according to the company.

The PDX9 warehouse has a reputation for having harsh working conditions; in 2018, an investigation from Reveal, an investigative journalism outlet, found that 26% of employees at the warehouse had sustained injuries. A report based on 2024 OSHA data showed that the company’s fulfillment centers report serious injuries at a rate more than two times the warehouse industry average.

Amazon’s fulfillment centers have been subject to several probes by federal agencies and prosecutors over warehouse safety, with investigators alleging that the company manipulated data and failed to properly document workplace injuries. The United States Attorney’s Office for the Southern District of New York is conducting an ongoing investigation into workplace safety at Amazon warehouses.

Amazon told TechCrunch that the company has seen a 43% reduction in its global recordable incident rate since 2019 — a metric that tracks any work-related injury requiring more than basic first aid. The company said it has invested more than $2.5 billion in safety improvements since 2019, including hundreds of millions of dollars in 2026 alone.