Robots equipped with Generalist AI’s new GEN-1 model have evolved into the ultimate automation workhorses, capable of completing the same simple tasks repeatedly without fail. You can witness in real time as one arm folds shirts row after row, carefully placing each one neatly into a basket, while another robot services robotic vacuums.

It’s only been a few days since Generalist AI introduced GEN-1 to the world, but this startup is already demonstrating how far AI can take robotics in a short period of time. Speaking of which, Generalist AI has only been active for two years, but the amazing work of its three creators gives us a sense of the enormous potential that awaits us here. There’s the one who worked on a key Google robotics project. Another was the creative mind who devised a completely new method of capturing data while observing someone pick up trash with a grabber. Then there’s the guy who worked for Boston Dynamics, the company that created the famously horrifying (but interesting) Big Dog robot. Together, they’ve gathered a lot of backing to pursue the idea that robots could handle the mundane activities of everyday life in the same way that flashy new chat systems handle language.

Unitree G1 Humanoid Robot（No Secondary Development）

• Sleek & Durable Design: Standing at 132cm tall and weighing only approx. 35kg, the G1 is constructed with aerospace-grade aluminum alloy and carbon…
• High Flexibility & Safe Movement: Boasting 23 joint degrees of freedom (6 per leg, 5 per arm), it offers an extensive range of motion. For safety, it…
• Smart Interaction & Connectivity: Powered by an 8-core high-performance CPU and equipped with a depth camera and 3D LiDAR. It supports Wi-Fi 6 and…

The key to all of this magic begins with some incredibly cool, and, let’s be honest, slightly strange, strap-on devices worn on their wrists. These effectively transform typical hands into miniature robot grippers, allowing you to perform a variety of normal tasks. Over time, this approach has accumulated an incredible five hundred thousand hours of genuine human interaction, and it’s all from people doing their thing in the real world.

Once that foundation is established, the engineers add a bit of robot-specific practice, as we’re only talking about one hour of robot motions each project, at most. Somehow, that’s all it takes to make this super-smart system that can see its surroundings, and then send out all the action commands it needs to get the job done, and to think, GEN-1 is built directly on top of the last model from just five months ago. It demonstrates how much further they’ve pushed the limits of what’s possible.

Let’s get to the figures, since that’s what matters in the end. For starters, dependability has improved dramatically. Their previous versions only got the job done approximately 64% of the time, but GEN-1 has increased that average success rate to a solid 99%. In one test, the robot worked nonstop for more than an hour, putting auto parts into kits without a single human hand lifting a finger, while in another, the robot slotted block after block into tidy little kits for an eye-watering 1800 successful cycles. When it came to packing phones, the robot surpassed 100 cycles, demonstrating an incredible level of precision and consistency.

This is significant because, at the end of the day, factories and warehouses are all about machines that simply keep working, day after day, hour after hour, without ever pausing to ask for a break. With GEN-1 at the helm, these facilities can finally rest assured that their machines will complete the task at hand, regardless of the circumstances.

Then there’s improvisation, which is the third element of the GEN-1 puzzle. When something unexpected happens, the robot simply adapts, making adjustments on the go. A washer is knocked out of place, and before you know it, the arm is reaching out with both grippers to nudge it back into place in a clever little trick. Or if a bag becomes caught in the middle of stuffing with soft toys, the machine just jiggles the bag briefly before sliding the item in without disturbance.
[Source]

Finalists for the prestigious EY Entrepreneur of the Year programme include Manna’s Bobby Healy, Neurent Medical’s Brian Shields and Sisterly’s three co-founders.

It is a list that always catches the eye. The EY Entrepreneur of the Year (EOY) programme finalists for 2026 have been announced today and there’re some very familiar faces to our readers. According to EOY, the 24 companies featured collectively employ more than 3,000 people, generating revenues of close to €1bn.

Hot on the heels of Manna’s $50m investment and jobs announcement yesterday (1 April), its founder Bobby Healy has earned his place in the ’emerging’ category, as has Brian Shields, whose Galway medtech start-up Neurent Medical closed an oversubscribed $62.5m Series C financing round for its Neuromark medical device to treat chronic rhinitis back in February.

It’s little surprise also to see the inclusion of Sisterly and its three co-founders Aoife Matthews, Jennifer O’Connell and Louise O’Riordan. The three seem to have captured the women’s health zeitgeist with their easy-to-use, science-backed ‘The Elevator’ supplement, which is designed specifically for women and was endorsed early by some big names like former star athlete Sonia O’Sullivan.

In total, there are 31 entrepreneurs representing 24 companies from across the island of Ireland shortlisted over three categories.

EOY Finalists – Emerging

• Aoife Matthews, Jennifer O’Connell & Louise O’Riordan of Sisterly
• Bobby Healy of Manna Air Delivery
• Brian Shields of Neurent Medical
• Danny Buckley of ADHDNow
• Dr Harriet Tracey of Beyondbmi
• Jennifer Rock of Skingredients Ltd
• Laura McCarthy of Drink Botanicals Ireland
• Liam Kearney of The Revive Group

EOY Finalists – Established

• Tommy Kearns of Xtremepush
• John Corley of Spanish Point Technologies
• John Lunn of Lunn’s
• Karl Zimmerer of Glanua
• Oltian Dervishi of Forte Pespa
• Paul Vallely and Clare Walsh of Kukoon
• Robin Jones of Golden Bake
• Trevor Casey of EPH Controls

EOY Finalists – International

• Brian Moloney of StormHarvester
• Chris Horgan of Dexgreen
• David Brennan of Eastgate Engineering
• Dermot O’Shea of Taoglas
• Ian Morgan and Sanjay Abraham of XPress Healthcare Ltd
• Maire Claire Reid of TST Group
• Paul, Gary & Mike Martin of Martin Group
• Ronan and Conor Burke of Inscribe

“Entrepreneurship in 2026 looks very different to even a few years ago,” said Roger Wallace, head of assurance & EY Entrepreneur of the Year, Ireland.

“We are operating in a world shaped by geopolitical instability, rapid advances in AI and a level of economic uncertainty that demands creativity as much as resilience. Our finalists this year have not only adapted to this environment, but are shaping it.”

This year’s finalists have been shortlisted by an independent judging panel of previous winners chaired by Harry Hughes, who is CEO of Portwest.

“Selecting just 24 finalists was an exceptionally difficult task,” Hughes said. “The entrepreneurs we met represent both long‑established sectors and fast‑emerging fields, but all of them are reshaping their industries with new ideas and new ways of working, and are turning challenges into opportunity.”

The EY Entrepreneur of the Year Ireland programme is supported by Julius Baer, Enterprise Ireland and Invest Northern Ireland.

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.

When Fizz quietly debuted in Saudi Arabia in the middle of March, founder and CEO Teddy Solomon wasn’t expecting the app to catch on like it did. Within 48 hours, the app hit No. 1 overall on the App Store charts, and users in the country have since sent more than 1 million messages. For an anonymous social app that started on college campuses, it was a surprising debut.

Solomon and his co-founder, Ashton Cofer, started Fizz in 2022 while they were students at Stanford, before dropping out. After raising $40 million and launching on 700 campuses, the app has been pushing beyond its college roots with Fizz Feed, a feature that opens the platform to non-students through location-based communities. Think of it as similar to Reddit, but without the ability to create or join topic-specific communities. Saudi Arabia — where Fizz currently holds the No. 1 spot in the news category — is its first overseas test of that ambition.

“We’ve always known that our big goal is to be a generational social product, rather than a college social app, and now we’re finally executing on it,” Solomon said. 

Fizz has not previously spoken about its international expansion.

Solomon said that when he attended a conference in Dubai, he saw the potential for Fizz’s expansion into the Middle East. Soon after, Fizz marketing analyst Michael Fonseca moved to Saudi Arabia to make connections in the area and better understand the culture, which paved the way for Fizz’s international launch.

“Mike was really welcomed with open arms,” Solomon said. “I think [Saudi Arabia] changed quite a bit in recent years.” The country is “jumping right now,” said Solomon. “Business is booming. The social scene and social landscape is booming. Snapchat’s huge there. And social apps are just massive in the region, whether it’s Snap, or WhatsApp, or TikTok — whatever other app it might be.”

This shift in the country’s image is intentional. In 2016, Crown Prince Mohammed bin Salman launched a government plan called Saudi Vision 2030, which aims to decrease the country’s financial dependence on oil. This strategy involves modernizing the country’s image — women can now legally drive, for instance — and investing in Western technology companies, like Google and Uber. More recently, the crown prince launched a state-funded AI company called Humain.

Despite these changes, Saudi Arabia remains an absolute monarchy, ruled by a royal family that suppresses free speech. In 2024, the Saudi government sentenced Manahel al-Otaibi to 11 years in prison for the “terrorist offense” of tweeting about women’s rights and posting photos on Snapchat in which she was not wearing a traditional abaya, according to Amnesty International.

Operating in Saudi Arabia, Fizz has to be aware that the monarchy could monitor its app for posts it deems offensive, demand that certain content be taken down, or even arrest someone based on their Fizz posts. Solomon doesn’t have a clear plan for how Fizz would handle such situations.

“The answer is, [we will] cross that bridge when we get there,” he said. “We have a lot of confidence in our guidelines. We are moderating very strictly and in a way that is satisfying people in the region and making sure that we’re abiding by the rules of the region and rules of the country.”

Solomon said that Fizz has invested heavily in Arabic natural language processing tools to support its content-moderation efforts. The company has also onboarded “hundreds” of volunteer moderators from the Saudi Arabian Fizz community. Fizz uses a similar strategy in its college communities — it uses AI content moderation tools, but it also seeks out volunteer moderators who have a better understanding of the nuances of campus culture, giving them more context when making moderation decisions.

Fizz says it has not received investment from any Saudi Arabian entities and has not communicated with any members of the government.

“There’s a lot of care for their community,” Solomon said. “There’s a lot of pride in their country, a lot of pride in the city that they live in, and they like the platform. They want to keep the platform safe, and they take a lot of honor in doing so.”

In short: Meta has suspended its collaboration with Mercor, a $10 billion AI data startup, after a supply chain attack exposed what may be the AI industry’s most closely guarded secrets: not just personal data, but the training methodologies that power the world’s leading large language models. The breach, carried out via a poisoned version of the LiteLLM open-source library, has triggered investigations at OpenAI and Anthropic, and resulted in a class action lawsuit affecting more than 40,000 people.

When hackers poisoned a widely used open-source library last month, they did not just steal personal data. According to reporting by Wired, they may have walked out with the blueprints for how some of the world’s most powerful AI models are built.

Meta has paused its work with Mercor, a San Francisco-based AI data company that generates bespoke training datasets for the biggest names in artificial intelligence, after a cyberattack exposed sensitive information about how the company, and potentially several of its other clients, actually trains its models. The pause is indefinite, and the incident has sent a ripple of anxiety through an industry that has spent billions developing the proprietary methods it was counting on keeping secret.

The startup behind the curtain

Mercor is not a household name, but it sits at a critical juncture of the AI economy. Founded in 2023 by Brendan Foody, Adarsh Hiremath, and Surya Midha, three Bay Area high school friends who competed together on the Bellarmine College Preparatory Speech and Debate team, the company recruits networks of human contractors, engineers, lawyers, doctors, bankers, and journalists, to produce high-quality, proprietary training data for AI labs. Its clients have included Meta, OpenAI, Anthropic, and Google.

The startup’s rise has been extraordinary even by Silicon Valley standards. In October 2025, Mercor closed a $350 million Series C round that valued it at $10 billion, minting all three founders as the world’s youngest self-made billionaires at the age of 22. By September 2025, the company had reached $500 million in annualised revenue, up from $100 million just six months earlier. Its business model, generating the fine-tuning and reinforcement learning data that AI labs rely on but rarely discuss publicly, made it one of the most valuable private companies in the AI supply chain.

That same positioning is now the source of its vulnerability.

A poisoned package, a cascade of exposure

The attack that reached Mercor originated several steps upstream. According to analysis by Wiz, Snyk, and Datadog Security Labs, a threat actor group known as TeamPCP compromised the CI/CD pipeline of LiteLLM, an open-source Python library used by millions of developers to connect applications to AI services, with 97 million monthly downloads and a presence in an estimated 36% of cloud environments.

TeamPCP had earlier used a supply chain attack on Trivy, a widely used security scanner, to obtain credentials belonging to a LiteLLM maintainer. On 27 March 2026, the group used those credentials to publish two malicious versions of the LiteLLM package, 1.82.7 and 1.82.8, directly to PyPI, the Python package repository. The tainted packages were available for roughly 40 minutes before being identified and removed.

The payload was sophisticated. Version 1.82.7 embedded base64-encoded malware directly into the library’s proxy server code, executing on import. Version 1.82.8 used a malicious path configuration file that triggered automatically on every Python process startup. Both variants were designed to harvest environment variables, API keys, SSH keys, cloud credentials across AWS, Google Cloud, and Azure, Kubernetes configurations, CI/CD secrets, and database credentials, exfiltrating everything to a server at models.litellm[.]cloud.

Mercor, which confirmed it was “one of thousands of companies” affected by the attack, subsequently found that the breach had exposed approximately four terabytes of data. According to court filings and claims made by the hacking groups involved, the stolen cache includes 939 gigabytes of platform source code, a 211-gigabyte user database, and roughly three terabytes of video interview recordings and identity verification documents. The exposed information may include the full names and Social Security numbers of more than 40,000 current and former Mercor contractors and customers.

The secrets that matter most

The personal data exposure would be troubling enough. But what has alarmed Meta and drawn the attention of other AI labs is a different category of information entirely.

Because Mercor sits inside the data pipelines of multiple AI companies simultaneously, the breach may have exposed details about data selection criteria, labeling protocols, and training strategies that companies have spent years and billions of dollars developing. Competitors can replicate a dataset; replicating a training methodology is harder, and it represents a genuine competitive moat. The Wired report notes that the scale of that potential exposure has prompted multiple AI labs to investigate what, precisely, may have left their orbit.

OpenAI, which also uses Mercor’s services, has said it is investigating the incident but has not paused its current projects with the company. Anthropic, which raised $3 billion in early 2026 and has been expanding its research infrastructure aggressively, has not publicly commented on its exposure. Google, which operates competing data vendor relationships of a similar kind, is also understood to be assessing the breach’s scope.

The incident illustrates a structural risk that the AI industry has rarely had to confront: when multiple competitors rely on the same third-party data supplier, a single breach can expose the competitive secrets of all of them at once.

Extortion and legal fallout

The threat group Lapsus$, which has previously been linked to high-profile attacks on major corporations, subsequently claimed responsibility for the Mercor breach and began auctioning the stolen data on dark web forums. Security researchers believe Lapsus$ is acting in collaboration with TeamPCP, which has emerged as a systematic threat across the AI and enterprise software ecosystem. The same group is believed responsible for a wave of supply chain compromises affecting more than 1,000 enterprise SaaS environments via the earlier Trivy attack, including a breach of the European Commission attributed by CERT-EU to the same campaign.

On 1 April 2026, plaintiff Lisa Gill, a resident of Wahiawa, Hawaii, filed a class action complaint against Mercor.io Corp. in the US District Court for the Northern District of California. The suit alleges that Mercor failed to maintain adequate cybersecurity protections, leaving more than 40,000 people exposed to identity theft and fraud. The complaint states that the LiteLLM incident on 27 March was the entry point and that Mercor’s reliance on a compromised open-source dependency without sufficient monitoring created the conditions for the breach.

Meta, meanwhile, has said nothing publicly, a silence that speaks volumes. The company signed a $27 billion AI infrastructure deal with Nebius Group in March 2026 and has forecast capital expenditures of between $115 billion and $135 billion for the year, making its AI training pipeline one of its most strategically sensitive assets. Pausing a data vendor relationship, even an important one, is the kind of decision that gets made only when the risk to proprietary methodology outweighs the operational cost of stopping work.

A cautionary tale for the AI supply chain

The Mercor breach is, in one sense, a conventional supply chain attack: a threat actor found a weak link in an open-source dependency and exploited it for credential theft and data exfiltration. In another sense, it is something newer and more unsettling. The AI industry has built its most valuable intellectual property on top of an interconnected web of data vendors, open-source tools, and shared infrastructure, and that web now constitutes an attack surface that no single company fully controls.

Security companies have been warning about precisely this dynamic. Aikido Security, which reached unicorn status in January 2026, built its business on the premise that open-source dependency risk had become existential for enterprise software. The Mercor incident suggests the same logic applies, perhaps more acutely, to the AI training pipeline.

For the three young founders who built one of the fastest-growing companies in tech, the coming months will test whether Mercor’s extraordinary momentum can survive a breach that exposed not just its users’ data, but its clients’ most carefully guarded secrets. The AI industry’s breakneck 2025 was built on the assumption that the infrastructure underpinning it was secure enough to trust. That assumption is now under review.

For the last couple of years, Microsoft has been all-in on Copilot. It’s literally everywhere, be it Windows, Edge, Office, or even baked into core workflows where you can’t really ignore it. The messaging has been clear: this is the future of productivity, your AI assistant for getting real work done.

And now, suddenly, Microsoft is saying… don’t take it too seriously.

Microsoft is walking back Copilot’s “serious use” pitch

As reported first by Tom’s Hardware, the Microsoft Copilot Terms of Use state that Copilot is intended for “entertainment purposes only” and shouldn’t be relied on for important or high-stakes decisions. That includes things like financial, legal, or medical advice. Basically, the kind of stuff people are increasingly using AI for.

Copilot is for entertainment purposes only. It can make mistakes, and it may not work as intended. Don’t rely on Copilot for important advice. Use Copilot at your own risk.

On paper, this makes sense. AI can hallucinate, get things wrong, and occasionally sound far more confident than it should. From a legal standpoint, this disclaimer is almost expected, as it acts like a safety net to avoid potential liability as these tools scale.

But here’s where it starts to feel a bit off. This is the same Copilot Microsoft has deeply integrated into Word, Excel, Outlook, and Teams. In fact, they’re even baked into Microsoft’s own Enterprise solutions, as pointed out by users. Tools that people use for actual work, not casual experimentation. When your AI is summarizing emails, drafting reports, or analyzing data, calling it “entertainment” feels oddly out of sync with reality.

The internet isn’t exactly buying it

Unsurprisingly, the internet isn’t exactly applauding. The reaction has mostly been confusion mixed with plenty of eye-rolls. Because let’s be honest, if Copilot isn’t meant for serious use, why is it sitting front and center inside tools people rely on to do serious work?

It’s starting to feel less like a redefinition and more like a safety net. Push Copilot everywhere, make it unavoidable, sell it as the future, and then quietly add a “don’t rely on it” label when things get complicated. It’s a neat way to enjoy the upside of AI while sidestepping the responsibility that comes with it.

Now, sure, Microsoft isn’t alone here. Every AI tool comes with some version of this disclaimer buried in the fine print. But most of those tools are optional. You install them, you try them out, and you decide how much to trust them. Unfortunately, Copilot did not follow that route. It showed up across Windows and Office and made itself part of the experience, whether you asked for it or not.