2025 was a major year for new 360 cameras. Insta360 maintained its dominance with the X5 (since upgraded with the X6), DJI entered the picture with the Osmo 360, and GoPro finally upgraded its original Max after years of delays with the Max 2.

In our in-depth review, we said GoPro’s 8K second-gen model was ‘worth the wait’, and it’s now available for a new low price that further sweetens the deal over Labor Day. You can grab the GoPro Max 2 at Amazon for $299 — that’s a full 40% off the original (already reasonable) $499.99 asking price.

Compare that deal to the current prices of rival models available in the US, which include the Osmo 360 and Insta360 X5 and X6, and the Max 2 is easily the best value 360 camera that we’ve tested and recommend. If you’re starting from scratch, the Max 2 Accessory Bundle at Adorama for $369.99 is an even better bet, adding a 1.2m extension rod, memory card and second battery.

We’re in turbulent times for 360 cameras; DJI is banned from releasing new models in the US (including this week’s Osmo 360 II), Insta360’s prices keep going up as seen with the recent X6, while GoPro’s financial decline was seemingly arrested this week after it ‘merged’ with an obscure brand that manufactures America’s AI infrastructure — but who knows what’ll happen with the action cam giant next.

A price cut for GoPro’s latest 360 camera is, therefore, some light relief. After all, the Max 2 shoots some of the best-looking 8K 360 footage in daylight, offers versatile mounting options, plus tool-free lens replacements for a cost-effective fix should the bulbous lenses get damaged in the line of action.

Sure, the Max 2 is prone to overheating and its image quality suffers a little in low light, but overall it remains a superb 360 camera, and is now available for an unbeatable price. If you want the best bang for your buck, look no further than the above Max 2 deals.

Not in the US? You can find the best GoPro Max 2 deals via the links below.

Shop more of today’s best Labor Day sales

• Amazon: 45% off TVs, AirPods, air fryers & vacuums
• Apple: iPads, AirPods & MacBooks from $99
• Appliances: up to $1,200 at Samsung
• Best Buy: 60% off appliances, TVs, laptops & more
• Casper: up to 30% off mattresses
• Dell: laptop deals from $249.99
• DreamCloud: up to 60% off mattresses, 66% off bundles
• Home Depot: 40% off appliances, furniture, grills & tools
• Lowe’s: $1,000 off appliances, patio & tools
• Nectar: up to 50% off mattresses, deals from $349
• Saatva: up to $625 off luxury mattresses
• Samsung: up to $2,000 off 4K, QLED & OLED TVs
• Target: 45% off clothing, tech & furniture
• Walmart: 58% off back-to-school, furniture, & clothing

Integrity360’s An Nguyen discusses the cyberthreats affecting industrial environments and what makes an effective OT security plan.

Operational technology (OT) has become a growing priority in the world of cybersecurity in recent years.

Securing OT systems, which operate and control physical processes in industrial environments while maintaining safety, reliability and operational continuity, is more important than ever amid the rising threat of disruption from cybercriminals and state-sponsored bad actors – who increasingly target OT systems in critical infrastructure in particular.

Sectors particularly affected by these types of cyberattacks include the likes of manufacturing, energy, transport, water and healthcare.

In fact, recent data from Statista found that the manufacturing industry was hit by the largest share of cyberattacks in 2025, with companies in this sector experiencing 27.7pc of total attacks for the year.

An Nguyen, OT practice lead at cybersecurity specialist Integrity360, tells SiliconRepublic.com that OT security considerations differ quite considerably from those of traditional IT security.

“Unlike traditional IT environments, where the primary objective is often the protection of information, industrial environments must balance cybersecurity with operational and safety requirements,” says Nguyen. “Effective OT cybersecurity requires not only cyber expertise, but also an understanding of industrial processes, operational dependencies and real-world constraints.”

Here, Nguyen discusses the concerns and complexities of OT security further.

What are some of the biggest threats facing OT environments?

Ransomware remains one of the biggest concerns, especially when attacks start in the corporate IT environments and then pivot to industrial operations. Critical infrastructure operators also face risks from sophisticated threat actors, supply chain compromises and insecure remote access.

However, what we often see is that successful attacks are not always the result of highly sophisticated techniques. They are frequently the result of more basic weaknesses such as poor asset visibility, weak network segmentation, legacy systems or uncontrolled third-party access.

As IT and OT become increasingly interconnected, organisations need to focus not only on external threats, but also to understand how disruptions to digital systems could propagate to critical industrial processes and services.

What are the biggest priorities in an efficient OT security plan?

An effective OT security strategy should start with understanding the business and operational environment. The first step is to identify critical processes, understand how they depend on digital systems and gain visibility over assets, communications and third-party connections.

Organisations also need a clear governance and operating model across IT and OT environments. One of the most common challenges is defining roles and responsibilities between operations, engineering, IT and cybersecurity teams. Without clear ownership, accountability and decision-making processes, even well-designed security programmes will face difficulties to provide sustainable improvements.

From there, organisations can assess their maturity level and build a roadmap based on risk and operational priorities. Typical focus areas include remote access security, network segmentation, change management practices, vulnerability management, third-party vendors assessment, monitoring, incident response and recovery capabilities.

The challenge is not to deploy as many security controls as possible. The challenge is to build a security model that takes into account the operational reality of the organisation and can be maintained over time.

Why is it important to keep OT security up to date?

Keeping OT security up to date is more than patching systems. It requires maintaining visibility of the environment, reassessing risks as the environment evolves and making sure preventive measures, detection and response capabilities remain effective.

This is particularly important because many industrial systems stay in operation for decades. You cannot always patch, upgrade or replace them at the same pace as you would in an IT environment. As connectivity increases and new vulnerabilities are discovered, organisations need to continuously adapt their security posture while managing operational constraints.

In what ways do companies fall short of proper OT security?

Most organisations do not fail because they ignore security completely. Usually, what we see is that security initiatives remain isolated, too much technology-focused or disconnected from operational realities.

Many organisations still struggle with basic visibility of OT assets, clear ownership between operations, engineering and cybersecurity teams, third-party access management and preparation for incidents that could impact production.

Another common challenge is moving from assessment to execution. Organisations often complete audits and assessments, but then struggle to prioritise, fund and implement the improvements that have been identified. Technology is important, but effective OT security also requires governance, operational ownership and long-term commitment.

What does a significant OT cybersecurity breach look like? Do you have any examples of a major incident?

A major OT cyber incident does not necessarily involve attackers taking direct control of industrial systems. In many cases, disrupting connected IT systems or remote access infrastructure is enough to affect industrial operations.

A recent example is the cyberattack that temporarily shut down a small power generation plant in the UK. According to public reporting, the incident is believed to be linked to Iranian actors and resulted in the facility being offline for several days. While there was no impact on the wider UK energy system, the event highlighted how cyberattacks can affect critical infrastructure operations without necessarily targeting industrial control systems directly.

This case illustrates that the challenge is not only to protect individual systems. It is to understand the dependencies between IT, OT and operational processes, and to make sure the organisation can continue operating and recover effectively when disruption occurs.

Is there a contrast between how organisations view IT security versus how they view OT security?

Historically, IT and OT security developed around different priorities. IT security has mainly focused on protecting information systems, while OT environments have traditionally prioritised safety, reliability and operational continuity. Today, the distinction is becoming less clear because IT and OT are increasingly connected, and OT is using progressively more common IT technologies.

The challenge is absolutely not to keep IT and OT security completely separated. It is neither to apply IT security practices to OT without adaptation. In practice, the most mature organisations combine both approaches. They use existing cybersecurity capabilities such as SOC, incident response and access management, while integrating OT expertise and operational context understanding.

Down the line, OT cybersecurity should be treated as a business and operational resilience challenge, not simply as an additional technical cybersecurity topic.

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.

personal tech

The FOSS tool for audio editing has a fresh coat of paint, and new features to boot

Audacity, a beloved open-source audio editing software tool, has just received a massive makeover.

For those unfamiliar with Audacity, it’s a 26-year-old piece of FOSS software that provides incredibly robust audio editing capabilities for the perfect price – $0 – making it ideal for independent musicians, garage bands, vinyl rippers, and college students who want to make a slowly, deliberately spoken foreign-language recording for a group project sound vaguely fluent – not that this vulture would know anything about that.

While an incredibly useful tool, Audacity’s interface has been stuck in the mid-aughts for a couple of decades now. That’s changed with Thursday’s release of Audacity 4.0.0. Not only has the interface been completely overhauled, but there are also a lot of under-the-hood changes as well.

The new UI was rebuilt on Qt and adds a Home screen for recent projects, plus configurable Workspaces. Audacity veterans will notice that the actual editing workflow looks different – much more like a modern piece of software than a relic of the Winamp days. 

As part of that redesigned interface, a number of new features bring Audacity into the current age: Clips can be moved over one another, with the underlying portion replaced where they overlap, multiple clips can be selected at the same time, and clips can be grouped for moving as a single object. Audacity 4 also retains non-destructive trimming, meaning trimmed portions of a clip can be recovered simply by extending the clip back into the trimmed-out portion.

Audacity users have been crying out for a dedicated splitting tool, parent company Muse Group said in a press release, and that’s been added in version 4. Clip envelopes have also been added, which allow users to adjust the volume and add fades to tracks without altering the original recordings. All of Audacity’s built-in effects have been given an interface redesign as well. 

In short, it’s a substantial rebuild, although Muse has tried not to alienate veterans, noting that “existing keyboard shortcuts remain available, projects created in Audacity 3 open normally, and the familiar layout of tracks and timeline has been preserved.” 

Audacity 4, like all the versions that preceded it in the past 26 years, is remaining open source and free as well. 

“We always like to thank the open source community who help our team make things better,” Muse’s head of communications Jack Sutton told The Register in an email. Sutton also tipped his hat to Martin Keary, former head of product for Muse Group, who was an instrumental part of building Audacity 4 and recently left the company. 

Speaking of Keary, he published a nearly hour-long YouTube video in October defending the early alpha build of Audacity 4 when the FOSS audio editing world maligned the early build for turning Audacity into a full-fledged digital audio workstation (DAW) used to record, edit, and produce audio, instead of the editing tool it had long been. Keary insisted that wasn’t the case, and Sutton reiterated the point to us. 

“DAW features like virtual instruments and MIDI support are not in Audacity 4,” the Muse Group spokesperson explained. “Some of the under-the-hood and structural changes do make it easier for our team to potentially build DAW-like features in future, but our current priority is on … speedy and efficient audio editing.” 

Audacity 4 can be downloaded on GitHub and on MuseHub for those who use Muse’s other tools. ®

Nearly a decade after Face ID debuted in the iPhone X, BASF is bringing a lawsuit against Apple for violating face authentication technology patents it holds.

Many versions of the iPad and nearly every iPhone since the iPhone X has used Face ID. For whatever reason, BASF chose today to pursue a patent infringement lawsuit.

According to a report from Reuters, the German-based BASF company has started a lawsuit against Apple. They allege patent infringement for their patents held on face authentication technology.

The company has a Houston location and has chosen Midland, Texas for the lawsuit. It’s a U.S. District Court known for patent lawsuit speed, given the relatively smaller criminal docket that the court processes as compared to surrounding regions.

Apple hasn’t shared a statement on the matter.

BASF is a corporate giant

It seems Reuters may have been tipped off on this lawsuit, so while we await an official filing, the information we can find doesn’t say much. There doesn’t seem to be any indication that BASF is a patent troll.

BASF is the largest chemical producer in the world. Though it does own many subsidiaries and technologies, some of which belong to smartphones.

TrinamiX is a subsidiary of BASF that was established in 2015. It holds over 370 patents and patent applications, and it has published work in facial recognition technology.

One such technology pertains to under-display facial authentication systems for OLED panels. It was shared in 2022 at a Snapdragon Summit and has shipped in Android smartphones.

The description of the technology certainly aligns with the rumored Apple implementation of under-display Face ID that is expected to debut with iPhone 18 Pro. However, it seems unlikely that the lawsuit could be referencing unreleased technology.

Once AppleInsider is able to review the lawsuit filing in full, more will be shared.

A prominent US senator is asking the National Security Agency to provide guidance to the general public on best practices for using virtual private networks to secure their communications from spying by foreign adversaries.

VPNs funnel all of a user’s Internet traffic through an encrypted connection to a remote server. The design provides strong assurances that no one between the user and the server can read the encrypted contents. VPNs also allow users to hide their IP addresses from the destination servers they communicate with. While US agencies have previously recommended use of VPNs, none have given recommendations on which ones provide adequate protection.

It’s all in the nuances

There are a host of limitations that can undo many of the protections users may think their VPN provides them. For instance, the encrypted tunnel often terminates once a single server decrypts the traffic and sends it on to its final destination. That means the decrypted traffic or the sending and destination IP addresses may be available for snooping by rogue employees or attackers who hack the server. VPNs also don’t encrypt certain types of metadata, such as time stamps, allowing nation-states to build profiles that can be useful in intelligence gathering.

With so many nuances, the existing recommendations to use a VPN don’t provide enough information for people to make informed decisions. Sen. Ron Wyden (D-Ore.) is asking the NSA to provide specific recommendations.

“Americans facing advanced foreign threats—including government personnel, defense contractors, journalists, and human rights defenders—deserve clear, honest advice about how best to protect their communications from surveillance by foreign adversaries,” Wyden wrote in a letter sent Wednesday to Gen. Joshua Rudd, the director of the NSA. “To that end, I request that you update NSA’s existing public guidance on VPN configurations to address this issue.”

Specific questions touch on some fairly technical details, including the general architecture of a VPN service. They include the adequacy of single-hop VPNs, which, as noted earlier, use a single server to decrypt traffic sent by the user and send it to its destination. It also asks about multi-hop architectures, in which the traffic is funneled through two or more servers, allowing the first to see only the IP address of the sender and the terminating server to see only the destination address. The letter also inquires about the use of random delays and cryptographic padding to thwart attacks that detect timing patterns or the size of messages. Wyden further asks about the adequacy of specific services such as Apple Private Relay, Nym, and Tor.

While OpenAI’s GPT-6 Astra is making a big splash in the AI world with its debut, it was far from the only one to be announced this week. When there’s a new, updated model seemingly every other day, it’s hard to stay up to date, let alone be excited, but the onslaught of new offerings is something to note. 

It wasn’t just OpenAI: Fellow AI heavy hitters Anthropic, Meta, and Google also announced updates to AI models. While each new model has its own strengths, a focus we’re seeing more and more often is advancements in agentic AI workflows, and each new announcement spotlight on those capabilities. 

Below, we’ll break down all of the latest models from this week and give you the details.

GPT-6 Astra

Of all models released this week, OpenAI’s is the only one that’s not a “.x” update. GPT-6 Astra is said to excel in cybersecurity and software engineering, and OpenAI calls it its most intelligent model yet. The performance benchmarks for Astra have it ahead of its competitors in a lot of ways, making it one of the most advanced models we’ve seen to date.

OpenAI is also touting Astra’s strength in multi-step agentic workflows, scientific discovery and financial modeling. The model is also said to be much better at computer use, which was a focus of its predecessor, ChatGPT 5.6, when it was released in July.

ChatGPT-6 Astra is rolling out to people with Daybreak access, its program for “vetted enterprise customers and cybersecurity practitioners.” OpenAI says that the new model will roll out to Plus, Pro and Business users over the coming days. 

(Disclosure: Ziff Davis, CNET’s parent company, in 2025 filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.)

Claude Fable 5.1 and Mythos 5.1

Anthropic announced updates for its Fable and Mythos models — both of which are some of the best available today. Anthropic says that both Fable 5.1 and Mythos 5.1 are the same model, but each has a different set of safeguards. Fable is the model that the average person can use, and Mythos is typically reserved for researchers and cybersecurity experts. 

Anthropic says its latest models have significantly better performance over their predecessors, and the release “sets a new standard” for coding, knowledge work and long-running problem-solving tasks. It also says that Fable 5.1, when set to low or medium effort, has performance similar to Fable 5 but at a much lower cost. 

Fable 5.1 is available to the general public, whereas Mythos is limited to Anthropic’s trusted access programs. 

Muse Spark 1.3

Meta released a model this week that should bring it back up into the conversation about the most capable models around. Muse Spark 1.3 brings more advanced reasoning and better performance in agentic and coding tasks. 

Meta said this model is trained to collaborate with the person using it, and will ask clarifying questions when a prompt is vague and check to confirm the actions it will perform in agentic workflows.

Must Spark 1.3 is available in Muse Code and Meta’s Model API.  

Gemini 3.8 and 3.8 Flash Cyber

Google released two new models in the form of Gemini 3.8 Flash and 3.8 Flash Cyber. This was hot on the heels of the release of Gemini 3.7 Flash on Aug. 13. The latest additions to the family are said to offer “next-generation intelligence” for agentic workflows and cybersecurity. 

Gemini 3.8 Flash is available to enterprise, consumer and developers across Google products, where 3.8 Flash Cyber is currently available in limited access to participants of the Fairwind Program, which was also announced the same day as the new models.

Blake has over a decade of experience writing for the web, with a focus on mobile phones, where he covered the smartphone boom of the 2010s and the broader tech scene. When he’s not in front of a keyboard, you’ll most likely find him playing video games or watching horror movies.

See full bio

Advertisement

Set 72306 PlayStation is LEGO’s first official collaboration with Sony, revealed September 3, 2026 during PlayStation’s State of Play. Builders 18 and up get 1,911 pieces that form a near 1:1 replica of the original grey console and its first wired controller. The finished set is just over 6 cm high, 26 cm wide, and 19 cm deep, which is close enough to the 1994 hardware to let the model to sit beside a real device without appearing toy-sized. The pricing in the United States is $179.99, while in the United Kingdom it is £139.99 and Europe gets it at €159.99.

The tactile Power and Open buttons function precisely as they did on the classic machine, while the lid on top opens similarly to the original disc holder. A short wire allows you to plug the pad into a front port, just like the original 1990s models. That first-generation controller has four small plastic faces: square, triangle, circle, and cross. However, because this is the launch console rather than one of the later DualShocks, the analog sticks are not included. The vents and button layout are reminiscent of the early SCPH-style design, rather than one of the later fancier upgrades.

When you open it, you’ll notice that the case contains two small diorama scenarios that can be removed. One side features a little mini circuit from the original Gran Turismo in 1997, with tiny little vehicles simply hanging around on the track. Opposite that, you get an Ape Escape scene from 1999, with the monkey and the light-up helmet, the red light swapping with the blue light on the tiny island layout. You may keep them inside the console or move them out so they sit right next to it. Speaking of which, Sony claims that the brickwork is filled of subtle allusions to the console’s own past.

LEGO Insiders can order on October 1, 2026. The set will go on sale beginning October 4 at LEGO.com, LEGO Stores, and select merchants; PlayStation will also offer it through its own store in some regions on October 1. The LEGO Builder app supplies 3D instructions so you can rotate the model while you work. For anyone who once waited for that disc tray to close, the finished grey box now opens, accepts a plugged-in pad, and still has two games waiting inside.

• Four federal agencies criticized for failing to share the framework’s details
• We don’t know anything about who’s involved and how models are tested
• Protect Democracy wants a court order by the end of the month

Four separate US federal agencies are being sued over the Trump administration’s secret framework for testing frontier AI models before they get released.

Protect Democracy argues that “almost no details” have been shared about how the framework works, who participates in it, how companies are selected or what legal authority underpins the reviews.

If you spend hours on calls every day, background noise can quickly become a problem, and Anker has a solution. At IFA 2026, the company introduced the Soundcore Space 2 Pro, the first over-ear headphones powered by its Thus AI chip.

Designed to improve voice clarity without sacrificing music quality, the new headphones combine AI-powered call processing, upgraded active noise cancellation, and long battery life in a package aimed at commuters, remote workers, and frequent travelers.

Anker’s Thus chip debuted back in April inside the Liberty 5 Pro earbuds, and Anker has since expanded it across its lineup, including new sleep earbuds that mask snoring and track your heart rate.

How Thus AI chip cleans up your calls

Space 2 Pro runs on Thus, Anker’s neural-net chip that fuses processing and memory together for a serious computing boost over its previous flagship earbud chip. During calls, a 6-microphone array works with Thus to separate your voice from everything happening around you, using AI trained on massive audio datasets instead of the older rule-based filtering most headphones rely on.

For noise cancellation specifically, Anker bumps that up to 8 microphones, running Adaptive ANC 3.0 and processing over 384,000 noise signals every second. Anker claims that’s 1.5 times stronger than what the Space One Pro delivered, with testing showing a 54.8% improvement in the 20Hz to 2,000Hz range – the exact frequencies covering voices, keystrokes, and general office chatter.

Beyond calls, HearID 5.0 builds a personalized sound profile by testing your hearing directly, and you get two EQ presets to choose from, along with LDAC support for higher resolution audio.

Soundcore Space 2 Pro specs, price, and availability

Battery life comes in at 60 hours with ANC off, or 30 hours with it running. Turn on LDAC alongside ANC, and that number drops to 25 hours. But only 5 minutes of charging buys you 8 more hours of listening.

The headphones also support spatial audio, multipoint connections, Bluetooth 6.1, and 20 built-in voice commands, plus wear detection, a feature the Space One Pro never had. These headphones will be available on September 22 for $200 in Midnight Grey, Jet Black, and Linen White across the US, EU, and UK.

In a late-night blog post last Friday, OpenAI said it would wind down its partnership with Cursor, the startup behind one of the most popular AI coding tools on the market. The reason OpenAI gave for the decision is effectively that it can’t trust Elon Musk, whose company SpaceX recently acquired Cursor in a $60 billion deal.

In doing so, OpenAI walked away from one of its most important customers. Cursor has long paid OpenAI fees to offer the ChatGPT-maker’s models to developers using its AI coding editor. At the start of 2026, Cursor was one of OpenAI’s top five customers in terms of revenue, people familiar with the matter tell WIRED. By the spring of this year, OpenAI estimated Cursor would bring in more than $1 billion in annualized revenue for the ChatGPT maker, based on the partnership’s performance at that time, the people said.

The figures, which have not been previously reported, indicate the financial loss OpenAI is willing to take on to avoid doing business with Musk. OpenAI declined to comment. Representatives for SpaceX and Cursor did not immediately respond to WIRED’s request for comment.

“We are making this choice because we cannot be confident that SpaceX will use our technology within our terms of service, based on our experience with Elon Musk’s companies violating contracts,” said OpenAI in the blog post.

As OpenAI prepares to go public next year, it’s trying to show investors a more stable business—ideally one that doesn’t rely so heavily on the goodwill of Musk. OpenAI reportedly now generates more than $40 billion in annualized revenue, drawing on several different lines of business, including subscriptions, ads in ChatGPT, and selling access to its AI coding tool, Codex.

Losing a major customer like Cursor may not hurt OpenAI financially as much as it once would have, but it was certainly a difficult decision. The company acknowledged in its blog post that ending the partnership may damage its standing with developers—a community it has spent years carefully working to win over—by cutting off a major channel they use to access OpenAI’s models.

Michael Truell, Cursor’s founder and CEO, who is now leading teams at SpaceX, responded publicly to OpenAI’s announcement hours later on Friday in a post on X, claiming that the ChatGPT-maker’s models only “serve about 5% of Cursor user traffic.” The comment seemed designed to imply that OpenAI’s models weren’t very popular with developers using Cursor, and wouldn’t meaningfully affect its business.

OpenAI’s head of core products, Thibault Sottiaux, was quick to point out that token usage is “not a proxy for revenue nor value created,” and asked Truell to “share the math” behind the 5 percent number.

WIRED previously reported that Cursor leaders hoped to remain a platform for third-party AI models from OpenAI and Anthropic, even after the SpaceX acquisition. Some AI insiders considered that wishful thinking, assuming that Cursor and OpenAI’s breakup has long been a foregone conclusion. As OpenAI builds Codex into a large AI coding business, it increasingly finds itself competing with Cursor for customers. Nevertheless, the companies have operated relatively peacefully, as partners and competitors, for well over a year.

Evidently, that arrangement was no longer possible with Musk in control of Cursor. OpenAI may have determined that its Cursor partnership was not worth the risk that SpaceX might distill its best AI models. The company noted in its blog post that, earlier this year, Musk seemingly said that xAI—now a part of SpaceX—used OpenAI’s models to train its own. The apparent admission came during Musk’s deposition in his lawsuit against OpenAI, in which he claimed that Sam Altman and Greg Brockman had stolen a charity they created together roughly a decade ago. A federal jury dismissed the suit earlier this year.