Got an M4 Mac? Nvidia’s new free tool lets it offload local AI tasks to your PC, turning your home network into a shared supercomputer.

The Personal AI Router (PAIR) isn’t hardware. Instead, it’s software that connects supported Apple Silicon Macs, Nvidia RTX PCs, and DGX Spark systems.

PAIR works with Ollama and LM Studio, two tools for running AI models locally. It discovers participating computers and gives AI apps a single connection for sending requests, so users don’t have to configure each app to reach every machine.

For someone already running local AI on a Mac, PAIR could put a compatible gaming PC to work when requests pile up. An AI agent reviewing several documents, for example, could have independent requests handled on different computers.

Nvidia released the open-source beta on September 3.

How PAIR distributes local AI requests

PAIR chooses an available computer for each request, while Ollama or LM Studio runs the model on that machine. When a computer is busy or unavailable, the router can direct new requests elsewhere.

The computers remain separate systems, according to Nvidia’s technical FAQ. PAIR doesn’t combine their GPUs or memory, so connecting two 16GB Macs doesn’t create one 32GB memory pool for a larger model.

Adding another computer doesn’t automatically make an individual AI response faster.

In a performance demonstration, Nvidia ran a task divided among five AI subagents in Hermes, using Ollama and the Qwen 3.6 35B A3B model. It reported average completion times of 18 minutes on an RTX Spark laptop and 8 minutes and 48 seconds across that laptop, a DGX Spark, and an RTX 5090.

Nvidia describes the result as an unofficial demonstration specific to that configuration. The comparison didn’t include a Mac, so it doesn’t establish how much a Mac user would gain from PAIR.

Mac requirements leave M3 Ultra outside the support list

Nvidia’s system requirements list macOS Tahoe and an M4 or newer chip for Mac support. The general requirements specify at least 8GB of RAM and recommend 20GB or more of disk space.

Features of Nvidia PAIR

The cutoff leaves the M3 Ultra Mac Studio outside PAIR’s published hardware requirements. Apple introduced that Mac in March 2025 with configurations offering up to 512GB of unified memory, and its launch announcement specifically promoted running large language models locally.

Nvidia’s requirements page doesn’t explain the M4 cutoff, and the published support list doesn’t establish whether PAIR would work on an older Mac.

Supported Nvidia hardware includes GeForce RTX 20-series GPUs and newer, RTX Pro workstation GPUs based on the Turing architecture or newer, and DGX Spark systems. PAIR supports Windows and Linux alongside macOS.

Users need to install PAIR, pair their computers, and make the required models available through Ollama or LM Studio. Each participating machine still needs the resources to run the requested model.

Nvidia describes PAIR’s local inference as keeping prompts, files, and agent context on the user’s network. Models need to be downloaded first, but PAIR itself doesn’t require an internet connection to operate.

The injection mechanism is a small device that deploys a syringe using a spring system. That triggers a process that breaks the seal between a chamber containing citric acid and another containing baking soda. The resulting chemical reaction—basically the same one you might’ve done as a kid to make a volcano with baking soda and vinegar—creates enough carbon dioxide to push the plunger on the syringe and administer medication.

Photograph: The University of Queensland

For the study, the research team remotely controlled a cockroach to travel from its starting point through three checkpoints and administer an injection to a simulated target.

The results showed that the success rate for injections made at close range—within 150 millimeters of the target—topped out at approximately 95 percent. The success rate for the entire sequence of tasks—from departure to completion of the injection—was 72 percent.

The researchers also successfully demonstrated teamwork, where one cockroach uses a camera to locate a simulated target while another serves as the injector.

Although the research team has previously developed cyborg beetles capable of climbing vertical walls, they note that larger cockroaches are better-suited for carrying specialized rescue and medical equipment.

While the results of the new research are promising, the experiment didn’t replicate the complex environmental conditions found at actual disaster sites, such as debris, uneven terrain, and shifting landscapes.

Vo Doan expressed hope that, if resources can be secured to speed up research and field testing, a rescue team of cyborg insects could be deployed at actual disaster sites within five to 10 years.

“Rather than building one robot to do everything, we can harness the natural strengths of different insects and equip them for different missions,” Vo-Doan says.

This story was originally published in WIRED Japan and has been translated from Japanese.

N-able has released an emergency hotfix for a maximum-severity remote code execution (RCE) flaw affecting its N-central remote monitoring and management (RMM) platform.

IT departments and managed service providers (MSPs) use the N-central platform to monitor, manage, and maintain client networks and devices from a centralized web-based console.

Tracked as CVE-2026-86218, this RCE vulnerability allows threat actors without privileges to execute malicious code on unpatched N-central instances exposed online in low-complexity attacks.

N-able addressed the flaw on Saturday by releasing N-central 2026.3 Hotfix 4 and urging customers to patch as soon as possible.

“At this time, we have no confirmations that this vulnerability has been exploited in production environments, but unpatched systems remain at risk,” the company said.

“Customers running on-premises N-central deployments should upgrade to N-central 2026.3 HF4 immediately to protect their environment.”

Internet security nonprofit Shadowserver Foundation now tracks nearly 1,500 N-central servers exposed online, most of them located in the United States and Europe.

Evidence of active exploitation​

While N-able has yet to confirm that the CVE-2026-86218 flaw is being targeted, cybersecurity company Huntress has flagged it as a potential zero-day, along with two high-severity vulnerabilities (tracked as CVE-2026-86206 and CVE-2026-86207, and also patched over the weekend) that can allow attackers to bypass authentication and gain full access to the vulnerable N-central platform.

“In our 9/5/26 update [..], we had said we could not rule out whether the two previous vulnerabilities released (CVE-2026-86206 and CVE-2026-86207) were the ones that were exploited in the instance seen in the patched production environment of one of our customers,” Huntress said.

“Because logs on the compromised N-central server had already rotated, we are also unable to say whether this new CVE was the vulnerability exploited in that case.”

“On-premises N-central users must apply HF4 immediately, as systems running HF3 remain vulnerable to this newly disclosed flaw,” Huntress warned.

One year ago, N-able released security updates for two N-central vulnerabilities (CVE-2025-8875 and CVE-2025-8876) that attackers were exploiting in the wild.

Days later, Shadowserver found that 880 N-central servers were still vulnerable to attacks exploiting the two security flaws even after CISA ordered federal agencies to patch their systems within a week and urged all security teams to also prioritize securing their systems against ongoing attacks.

Overall prevention scores can hide what happens after initial access. Once attackers are using valid credentials, prevention drops sharply.

The Blue Report 2026 measures defenses technique by technique across 338 million simulations run in customer production environments.

Get the report

The EU’s forthcoming Quantum Act will not come with a dedicated budget.  “The act is not about budget,” said Thomas Skordas, deputy director-general of the European Commission’s DG CNECT, at a European Parliament event last week, on 3rd of September.

“What the act will provide is the stimulus to make sure that we move in the right direction, that we bring people together.”

The session, hosted by the European Quantum Flagship and German Green MEP Sergey Lagodinsky, brought together representatives from the Commission, Council and Parliament, along with researchers and quantum company founders. The Commission hopes to present the legislation by the end of 2026, although that timeline could slip into next year.

The ambition behind it is significant. The EU has declared its aim to become the world’s quantum valley, and the act is intended to help make that happen.

Skordas described the act as an enabling framework rather than a regulation. That distinction matters in Brussels, where recent technology policies such as the AI Act and the Digital Markets Act have largely focused on imposing rules and obligations rather than bringing different parts of an industry together.

The clearest comment of the day concerned Europe’s quantum startups.

“We cannot afford having 78 startups all on their own,” Skordas said. “We don’t have Microsoft, we don’t have Google, we don’t have IBM. We want competition, but at some stage, we expect a little bit of consolidation.”

It is unusual for a Commission official to openly call for consolidation in an industry. But the comment also reflects a problem that has become increasingly difficult for Europe to ignore: its quantum efforts are spread across member states and different technology approaches, producing plenty of prototypes but fewer companies with the scale to build their own manufacturing capacity.

Tommaso Calarco, secretary of the EU’s high-level advisory board on quantum technologies, put it more simply.

“What do we need? We need critical mass,” he said. “And critical mass means three things: it means cash, it means speed, and it means we do things together.”

The act can address two of those three things. Cash is not one of them. Calarco was blunt about what he thinks is at stake: “This quantum act is the last train which is passing for Europe.”

He was also critical of how member states are working with the Commission. They “are not yet pooling their resources strategically to the extent that they should, and we cannot wait”, he said.

Europe does have one advantage that the usual argument about fragmentation can overlook.

“We’re not only buyers, but we are also suppliers,” said Cecile Perrault, executive director of the European Quantum Industry Consortium. “80% of the components that we find in quantum computers in Europe come from Europe. We have a huge advantage compared to other digital technologies there.”

That position in the supply chain is already attracting private investment. The Novo Nordisk Foundation is building an open quantum chip foundry in Copenhagen, creating the kind of manufacturing capacity the act is intended to encourage, and doing so without waiting for the legislation.

Skordas also argued that quantum computing needs to become part of a much broader technology ecosystem.

“We need to break silos,” he said. “Quantum has to be combined with semiconductors, photonics, the cloud industry, the end users and many other technologies that are more mature and can win new markets.”

Talent was another concern raised during the discussion. “The talent goes where the global leaders are, not only where the money is,” said Eleni Diamanti, co-founder of Welinq and a CNRS research director at Sorbonne University. That makes the talent problem closely tied to Europe’s ability to build companies that can become global leaders, rather than a separate issue.

Lagodinsky placed the debate in a broader context.

“It’s not just a competition between business models, it’s a competition between societal models,” the MEP said, “and, in this competition, technology and leadership in the realm of innovation play the role.”

Europe has already been investing significant funds in quantum technology, with public investment reported to be outpacing that of the United States, and France committing €500m of its own.

The bigger question is whether that investment can produce an industry capable of competing at scale, rather than primarily supporting research. That is the problem the Quantum Act will inherit.

For now, it is still unclear what the act will actually contain. There is no draft text, no fixed date for its presentation, and no details yet on how the Commission plans to encourage the kind of consolidation Thomas Skordas described.

on-prem

Couldn’t stop before the end of a runway

Two runways are currently closed at Miami International Airport after an Amazon-branded cargo plane overran a runway, resulting in five fatalities.

The leased 767, which bears the “Prime Air” brand Amazon uses for its air freight and drone operations, landed but was unable to stop before the end of the 2850-meter runway 30. The plane appears to have exited the runway and the airport grounds, crossed a road, and come to a halt in what looks like a carpark.

As it ran across grass beyond the runway, the plane’s nose pitched down into the ground, its starboard engine caught fire, and the craft sustained extensive damage.

Local authorities have confirmed that five people died, and three are severely injured.

A statement from Amazon spokesperson Kelly Nantel says the company is”heartbroken to learn that five people lost their lives in today’s incident at Miami International Airport.”

The statement identifies the operator of the plane as 21Air, a company that leases cargo planes and crew to its customers.

21Air CEO Keith Winters said “We are devastated by the accident involving one of our aircraft in Miami today. Our deepest condolences are with the families and loved ones of those who lost their lives. Our immediate priorities are supporting those affected, assisting the authorities, and ensuring that accurate information is communicated as it becomes available.”

At the time of writing, the crash is just over 12 hours in the past, so investigators’ work has just begun. A video purportedly showing the plane landing depicts dark storm clouds and rain around the airport. That weather could conceivably have created gusty conditions that made landing unusually challenging.

Federal records indicate the plane rolled off the Boeing production line in 1994. According to Planespotters, its first owner was Belgian carrier Sobelair, and it was later bought by carriers from Vietnam, Spain, Kenya, and  Russia, before freight airline Atlas acquired it in 2015 and converted it from passenger to freight operations.

That’s a common fate for older planes, which passenger airlines sell at very low prices. Cargo airlines can therefore pick up well-maintained planes at very low prices, meaning they have lower borrowing or leasing costs. That cheap purchase price offsets the higher maintenance and operating costs that come with older planes.

All commercial aircraft, however, share strict airworthiness requirements.

This is the second major crash involving a 767 leased by Amazon, after the 2019 incident that saw a plane unexpectedly nose-dive. ®