Hark, the secretive AI startup founded earlier this year by serial entrepreneur and roboticist Brett Adcock, today announced Handoff, a “computer use agent” (CUA) that it says is among the top-performing in the world at navigating the open web on a user’s behalf — ordering dinner on DoorDash, booking flights on United and Delta, or messaging job candidates on LinkedIn — all autonomously, end-to-end.

Sign-ups open to the public today at hark.com, with availability planned for later this month as part of the initial release of Hark’s software platform.

The company says Handoff recorded the top-ever score on Online-Mind2Web (OM2W), a third-party benchmark with a human-evaluated leaderboard for web agents, posting a 97.7 against 92.8 for OpenAI’s GPT 5.4, 84.1 for Anthropic’s Claude Opus 4.8, and 69 for Google’s Gemini 2.5 Pro.

Hark also says it can serve the model at less than one-tenth the token price of competing frontier models — $0.18 per million input tokens and $2.37 per million output tokens, versus $5 and $30 for GPT 5.5 — with per-turn model latency of 0.8 seconds.

For each request, Handoff spins up a dedicated virtual computer with its own browser, file system, and terminal, and users can connect existing accounts so the agent can log in and act with their saved addresses, payment methods, and history.

Hark’s research uncovered that despite people spending 75% of their screentime every day in a browser, fewer than 1 in 1000 websites have publicly accessible APIs, making it challenging for AI agents to take over the workload.

In a roughly four-minute produced announcement video posted on YouTube and social media, Adcock — seated in a bare warehouse space that doubles as a metaphor for the company’s build-out — speaks a request aloud to Hark (“let’s liven this place up a bit… let’s do some roses, maybe some cherry blossoms”) and Handoff is shown navigating a florist’s website to place the order, while Adcock narrates that unlike a typical chatbot, Handoff “is always working, it’s looping,” and says he now uses it for “all of my recruiting efforts end to end.” In Hark’s announcement blog post, more demos are shown in realtime and 5x speed.

But big some open questions about Handoff remain, especially for potential enterprise customers and users.

High-scoring benchmarks…but against last generation’s models

Notably, the benchmark comparisons Hark provided to VentureBeat for its Handoff AI agent are against GPT 5.5, GPT 5.4, Opus 4.8, and Gemini 2.5 Pro — the prior generation of frontier models.

The current leaders, OpenAI’s GPT-5.6 and Anthropic’s Opus 5, are absent, as are strong open-source computer-use contenders like DeepSeek V4, Kimi K3, and Qwen3.8-Max.

These newer models haven’t published Online-Mind2Web results, and no third party has posted them to the benchmark’s public leaderboard — meaning Hark’s “top-ever” claim cannot currently be checked against the strongest available systems.

The omission is notable because the newest frontier models have posted their largest gains precisely in computer use: on OSWorld 2.0, a related benchmark covering full computer control, Anthropic’s Opus 5 scores roughly 70.6% versus 55.7% for the Opus 4.8 model Hark chose as its comparison point.

The latency comparison comes with similar caveats: the 6.8-second and 6-second per-turn figures Hark cites for GPT 5.5 and Opus 4.8 were measured by Hark, in Hark’s own harness, with the competing models set to their highest — and slowest — reasoning level. No independent latency measurements exist for comparison.

Asked by VentureBeat whether Hark plans to publish comparisons against those newer models, the company did not specify.

Even within Hark’s own chosen comparisons, the “best” framing has an asterisk: on WebTailBench v2, one of the three benchmarks in Hark’s own results table, GPT 5.5 scores 72.3 to Handoff’s 68.6.

Two of the three benchmarks (WebTailBench and an unnamed internal evaluation) were also run inside Hark’s own harness, with pass rates computed by Hark’s internal LLM judge — conditions the company controls.

Hark’s pricing advantage is far clearer: Anthropic’s newer Opus 5 carries the same $5-per-million-input and $25-per-million-output list price as its predecessor, so Handoff’s roughly tenfold cost savings would hold up even against the current frontier — assuming its benchmark performance does too.

Training and file access

Hark’s research preview describes a sensible-sounding pipeline — supervised fine-tuning followed by asynchronous reinforcement learning using the GRPO algorithm, according to materials shared with VentureBeat prior to today’s announcement — but the company acknowledges it has only done post-training so far, with pre-training “planned for later this year.”

That means Handoff is built on top of a base model Hark did not train. Asked which base model it is, and what mix of proprietary and open data Handoff was trained on, Hark hasn’t yet specified.

Another big question mark for enterprise users: who can access the dedicated virtual computers and the files created on them?

A Hark spokesperson said “security and privacy is a primary focus, but this is a technical preview,” adding the company will share more when the product reaches market at the end of the summer.

Adcock’s history leading up to Hark

Hark is Adcock’s fourth company. He previously co-founded the talent marketplace Vettery (sold in 2018 for roughly $100 million), the air-taxi maker Archer Aviation, and the humanoid robotics unicorn Figure AI.

Hark raised a $700 million Series A round in May 2026 at a $6 billion valuation — led by Parkway Venture Capital, with participation from Nvidia, AMD, Intel Capital, Qualcomm Ventures, Salesforce Ventures, and ARK Invest.

Adcock seeded the company with $100 million of his own money and remains founder and CEO of both Figure and Hark simultaneously, a spokesperson confirmed.

Asked how the two companies interact, the spokesperson said Hark models “are being trained on the Figure robots,” but that Adcock has no plans to combine them.

Adcock’s promotional style has drawn skeptics. In April 2025, Fortune correspondent Jason Del Rey reported that Figure’s much-touted BMW partnership was far more modest than Adcock’s public claims of a robot “fleet” performing “end-to-end operations”: BMW spokesperson Steve Wilson said a single Figure robot was practicing picking up parts during non-production hours.

But the partnership has advanced, and as of June 2026, BMW said the Figure 02 robot supported production of more than 30,000 BMW X3 vehicles during a 10 month-period, and that the next-generation Figure 03 robot was being deployed at the plant for a parts-sequencing role in logistics.

On the social network X, Adcock called the story “mischaracterizations and downright lies” and threatened a defamation suit. Two months later, TechCrunch reported that Adcock skipped a promised live demo at a tech conference and sidestepped questions about the BMW deal onstage.

None of that means Handoff’s numbers are wrong. The agent may well be excellent, and the pricing — if it holds — would undercut every major lab.

The applications are in. The TechCrunch Startup Battlefield team made their decisions. Eight Australian startups have earned their spot on the Startup Battlefield stage at Stripe Tour Sydney on August 19 — and they’re ready to pitch for $15,000 in Stripe fee credits, investor attention, and a spot in Startup Battlefield 200 at TechCrunch Disrupt in San Francisco.

They represent the future of Australian tech — from enterprise tech to health and wellness tech to entertainment and media. Here’s who they are.

The eight finalists

Aigentsphere

Independent AI management and governance platform Aigentsphere helps enterprises scale AI safely and well, giving leaders visibility, control, and accountability over their entire AI agent workforce.

Apate.ai

Apate help banks, telcos, and governments dismantle scam operations by engaging scammers in live conversation and turning it into real-time intelligence.

Callease Ai

Callease Ai is the operating system for physical security control rooms, starting with voice AI that automates welfare, escalation, patrol, and many more — the entire calling engine.

Choosey App

Choosey App helps businesses hire instantly and workers get paid instantly through agentic AI-powered matching, workforce management, and same-day pay.

Doomers AI

Doomers AI helps AI and tech companies make product launches trend on X by orchestrating a vetted creator network that reaches the buyers who block ads.

equ

Startup equ helps health and wellness businesses deliver personalized nutrition at scale by embedding equ’s AI engine, built on 10 years of data and 100,000+ users, into their product.

LeadStory

LeadStory is the video intelligence layer powering the next generation of screens, surfacing exact video moments from sports, news, and finance clips to answer natural language queries.

Preve

Preve helps physical therapy clinics improve patient retention and clinical outcomes by automating the treatment plan creation and adherence.

Three will win prizes. One will go to San Francisco. All eight will pitch in front of investors, press, and the Australian tech community. If you want to see the future of Australian startups unfold, register for Stripe Tour Sydney on August 19 →

The stage is set. The judges are ready. Let the pitching begin.

Hosted and MC’d by Isabelle Johannessen, Director of Startup Battlefield.

• The Samsung Galaxy Fit 4 just leaked in FCC regulatory filings
• The documents suggest it will have a screen, unlike the Fitbit Air
• But the leak reveals previous few other details

The Fitbit Air has proved that you don’t need to pay a high price tag to get a great wearable, and now it looks like Samsung is getting in on the act with an alternative of its own. And with more details leaking out, we’re getting a clearer picture of the upcoming device than ever — but plenty of unanswered questions remain.

New FCC regulatory documents (spotted by Gadgets & Wearables) have spilled the beans on the much-rumored Samsung Galaxy Fit 4. There, you’ll find various nuggets of info on what we can expect from the product, as well as a few things that it likely won’t do.

At a time when storage space commands a premium price, SanDisk’s Extreme Fit drive is worth a look.

What is SanDisk Extreme Fit?

SanDisk

How fast is the SanDisk Extreme Fit USB-C?

SanDisk

Is the SanDisk Extreme Fit USB-C compatible with iPhones?

DenPhotos/Shutterstock

The U.S. Cybersecurity and Infrastructure Security Agency is giving federal agencies three days to mitigate vulnerabilities in IBM Langflow, N-central, and Apache Tomcat, all actively exploited.

​Tracked as CVE-2026-9198, the security issue in IBM’s Langflow visual framework for building AI agents is the most severe, with a critical rating of 9.8 out of 10.

It allows an unauthenticated attacker to execute remotely on default Langflow deployments by chaining two API endpoints to bypass login and run code.

In late July, multiple fully functional proof-of-concept (PoC) exploits for CVE-2026-9198 emerged in the public space, with complete instructions on how they can be leveraged.

Two weeks ago, CISA issued an alert for another critical Langflow vulnerability (CVE-2026-0770) being exploited in attacks to gain remote code execution with root privileges.

The vulnerability in N-able’s remote monitoring and management platform N-central is identified as CVE-2026-18576 and allows attackers to hijack administrative accounts without authentication.

The flaw received a high-severity rating and has been patched by the vendor. However, the fix was insufficient, and threat actors found a new way to exploit it.

N-able warned customers on August 1st that hackers were actively exploiting the new vulnerability, which received the identifier CVE-2026-18576.

An emergency hotfix was released on Sunday. The company urged customers to install it as the flaw impacted all versions of N-central before 2026.3.

The Apache Tomcat vulnerability, tracked as CVE-2026-34486, has a high-severity score of 7.5. It stems from an incomplete fix for CVE-2026-29146, a critical vulnerability with a severity rating of 9.8 that is described as the missing encryption of sensitive data.

On July 30, researchers at Palo Alto Networks Unit 42 reported that a Chinese-speaking threat actor tried to exploit the CVE-2026-34486 vulnerability in a manual campaign to plant reverse shells on nine Apache Tomcat servers.

CISA confirmed that threat actors are leveraging all three flaws in attacks and added them to its catalog of Known Exploited Vulnerabilities (KEV).

However, the agency did not share what types of attacks are leveraging them, noting that it is unknown if they are used in ransomware campaigns.

CISA has ordered federal agencies to apply available mitigations for the three targeted products by the end of Friday, July 7th.

Security teams log 54% of successful attacks and alert on just 14%. The rest move through your environment unseen.

The Picus whitepaper shows how breach and attack simulation tests your SIEM and EDR rules so threats stop slipping by detection.

Get the whitepaper

On July 6, at the dawn of a new fiscal year, Microsoft announced it was laying off 3,200 people in its Xbox division, plus an additional 3,200 employees across other departments. Of the Xbox layoffs, 1,600 jobs were cut immediately, while the remaining 1,600 will be lost over the coming year, the promise casting a morbid shadow over the immediate future of Microsoft’s gaming business. Additionally, Xbox attempted to shutter five of its development teams — Arkane Studios, Compulsion Games, Double Fine, Ninja Theory and Undead Labs — and it ended up divesting four of them, while the future of Arkane remains uncertain.

The layoffs hit Xbox at all levels: Doom creator id Software lost a reported 50 percent of its staff, Double Fine had to drop a third of its employees as it bought itself back from Microsoft, the platform team was gutted, and it appears that multiple projects were canceled at Fallout: New Vegas studio Obsidian Entertainment, including a sequel to Avowed.

Knowing it was coming didn’t ease the carnage.

Xbox CEO Asha Sharma and chief content officer Matt Booty, both of whom assumed their roles early this year in a surprise c-suite shakeup, outlined the brand’s dire financial situation in a June public memo. They highlighted broad internal issues with Xbox games, consoles, infrastructure and marketing, and ominously promised a “reset” for the business. Rumors of incoming mass layoffs and reports of potential studio closures swiftly followed, with the culling scheduled to begin just after Microsoft’s new financial year on July 1.

“We will end this fiscal year at about a 3 percent accountability margin, down year-over-year,” Sharma and Booty wrote in their memo. “Excluding Activision Blizzard King, over the past five years, we have spent over $20 billion on ongoing investments in our content, platform, and hardware subsidy, but our annual revenue has declined nearly half a billion during that time. Going forward, this cannot continue.” (Including Activision Blizzard King, which Xbox acquired in 2023, adds $69 billion to the company’s expenditures.)

Ahead of the layoffs, Xbox union members represented by CWA urged the company to consider the human and creative cost of repeated mass firings, and to negotiate in good faith over worker protections. Union members accused Microsoft of leaving proposals on the table for months at a time, paying little attention to meetings and unjustly allocating the company’s vast financial resources. UVW-CWA treasurer Sherveen Uduwana noted that Microsoft CEO Satya Nadella personally made $96 million in 2025.

“There is no shortage of wealth in the games industry, especially if we’re talking about Xbox, Sony, EA,” Uduwana said.

After Xbox announced the layoffs on July 6, the CWA in the United States and Canada filed lawsuits against Microsoft over unfair labor practices, and employees held protests outside their offices calling for “people over profits.”

Additionally in July, Xbox’s network went down for nearly a day and exposed that even physical media isn’t immune to DRM on the company’s consoles, though the disc issue was apparently a bug that will be fixed. Days later, Microsoft’s Q4 2026 financial results revealed a revenue decline of 10 percent for Xbox. Annually, Xbox’s reported revenue fell by more than $2 billion year-to-year, landing behind its closest competitor, PlayStation, by multiple billions.

All of this followed two years of rolling layoffs at Xbox, eight years of increasingly reckless studio acquisitions, and a systemic pattern of valuing potential revenue over any sort of creative process. But we’ll get to that at the end. First, an overview of Xbox’s terrible, horrible, no good, very bad July.

At first, the team envisions that those ideas will be codeveloped with humans, but their goal is deeply automating the process. If it works out, they contend, it’s possible that a few dedicated individuals may use Discovery Loop to out-invent the world’s largest research organizations. Presumably those small teams will be customers of Discovery Loop. (The startup also may make its own loops to seek and monetize breakthroughs.)

They put together a simple pitch deck and set out to talk to VCs about funding their company, which they set up as a public benefit corporation. “We just put a few slides with our backgrounds and a rough sketch of how we think our approach will be successful in a bunch of different ways,” says Dean. (They didn’t use AI to create it.) But their presence spoke louder than any PowerPoint could.

“With this team, I wouldn’t need to know what they were doing before I backed them,” says venture capitalist Vinod Khosla, who met with them in his Sand Hill Road office on a Saturday, so no one would get a hint of Google’s upcoming loss. “It’s the ultimate superstar team.” Khosla also loved their idea: “Humans have been using AI to do research, not using AI to be a researcher,” he says. “The fundamental thing [in Discovery Loop] is that AI is the researcher.”

Other meetings took place at Dean’s house. Ultimately Khosla Ventures and an AI-oriented VC firm called Radical Ventures bought in, along with a handful of other firms. (The founders aren’t sharing how much the funding was and at what valuation–but consider that some companies are paying tens or even hundreds of millions for a single AI superstar, and these guys are GOATs.) “Building something to solve all those problems is a powerful idea,” says Radical managing partner Jordan Jacobs, who will join the startup’s board. “These people have been doing this kind of work in the past so they know what they’re doing.”

Negotiating the exit from Google was more painful. Dean says that Alphabet CEO Sundar Pichai tried over multiple meetings to get them to keep their badges. Ultimately the team decided that they wanted the fun–and the freedom–of doing a startup.

“In a large organization there is always a lot of inertia you have to overcome to make any radical changes,” says Vinyals. “We want to build something different.” Google ultimately gave the departing engineers its blessing, along with an investment and an arrangement to provide compute power for the first year.

“Over 27 years, Jeff and Sanjay helped to drive some of the most significant technology transitions, from our early search infrastructure to the neural networks that helped create the modern AI era,” Pichai said in a statement. “We’ll continue to work with Discovery Loop as a founding investor and Cloud partner, and collaborate on a research framework for ML systems and related infrastructure advances.”

In part because of the impact this mini-exodus will have on Google. Dean and his cofounders have taken pains to maintain stealth. To date, they haven’t yet begun to hire a team (which may mean even more defections from Google) or even to rent office space. When I ask who the CEO is, there is a brief pause. “I think I’m the CEO,” says Dean, sheepishly. “Everyone pointed at me.”

Discovery Loop’s mission is not novel. Just about every major figure in AI—Dario, Sam, Jenson—has claimed at one point or another that AI will become an engine of scientific discovery, solving problems that humans could never crack on their own. This team is betting that their knowledge and skills will prevail in building a system that actually makes such breakthroughs. Short of that, Discovery Loop faces a narrower mark for success: Will the work they do as a startup be as significant as the talent gap they’re creating by leaving Google?

More than 200 people registered for the competition within the first eight days and applications will remain open until 9 August.

TechIreland has announced dates for the free annual National AI Challenge. Participants will engage with two intensive weeks of remote development, with teams presenting their work via coordinated regional events on 14 September. 

The strongest competitors will continue on to the TechIreland National AI Meet to be held at Galway’s Galmont Hotel on Thursday 24 September. Their work and knowhow  will be judged by leaders across technology, industry, investment, research and the public sector.

The 300 or so participants, from across the entirety of the island, will be made up of developers, founders, students, technologists and AI enthusiasts and will build practical AI products, prototypes and agentic workflows, with the aim of addressing real business and societal challenges. 

The competition is supported by expert mentors, industry partners, technology providers and more than 10 regional innovation hubs. More than 200 people registered within the first eight days and applications will remain open until 9 August.

This year marks the third iteration of the National AI Challenge. Last year, participants delivered 44 AI projects, with 15 teams reaching the national final – more than twice the number in 2024.

The theme for 2026 is ‘AI and Agents: Strategy, Processes and Roles’, which reflects the shift from AI experimentation to practical implementation, according to TechIreland.

Teams will be expected to respond to industry-led ‘challenge statements’ or submit their own wildcard ideas in areas such as enterprise technology, manufacturing, financial services, insurance, health, sustainability, public services and immersive technology. TechIreland said there will be an emphasis on responsible, people-centred AI that takes into consideration how AI is governed and used.  

Commenting on the announcement, Niall Norton, the CEO for TechIreland, said: “The AI Challenge is much more than a competition. It is becoming part of Ireland’s AI talent infrastructure. At a time when employers need evidence of practical AI capability, we are creating opportunities for people to demonstrate what they can build while helping organisations discover the talent they need.”

Brian Caulfield, board director at TechIreland, added: “I’m very proud to see the National AI Challenge and Meet go from strength-to-strength.

“Every year, the competition has generated more AI start-ups, while simultaneously educating new AI talent and connecting technical and commercial founders. I look forward to this year’s iteration being the best yet.”

The AI ecosystem has experienced a boom in Ireland, with many organisations and institutions focusing on education, scale and growth. In late July, the Tech Industry Alliance and RDI Hub teamed up to launch the Graduate AI Launchpad pilot, a new programme designed to prepare graduates, employers and industry professionals in Ireland’s south-west for AI-enabled workplaces.

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.