Embattled surveillance technology company Flock Safety unveiled a “generous” severance package for voluntary employee departures on Friday, according to a report in Wired.

Flock reportedly expects a significant portion of its 1,500-person workforce to express interest in the buyouts, and said it will grant them to a majority of those who are interested. The company’s internal announcement described these packages as the “most generous” it has ever offered.

By letting employees depart voluntarily, Flock can say goodbye to team members demoralized by the ongoing backlash over the company’s license plate recognition technology. Wired also reports that without buyouts, the company would “almost certainly” need to lay off some staff.

In August, The Washington Post identified 46 cases where police officers have been accused of misusing Flock technology, including cases where they allegedly stalked their wives, girlfriends, or exes. Florida and Texas both said they will stop using the startup’s technology, and an anti-surveillance advocacy group identified 90 cities that dropped Flock in August alone — a fourfold increase from the previous month.

TechCrunch has reached out to Flock for comment. The startup’s CEO Garrett Langley recently told the All-In podcast that the “biggest damage” caused by the backlash has been to “internal morale.”

The big picture: Researchers in Hong Kong have developed a technique that uses injected radio signals to extract audio and other information from headphones, phones and smart-home devices. Called InjectEave, the method targets analog components that can leak signals too weak to capture through conventional electromagnetic eavesdropping. In testing, the researchers recovered understandable headphone audio from up to 30 meters away, including through walls.

The research comes from the Hong Kong University of Science and Technology in Guangzhou and the Hong Kong Polytechnic University. The team presented its paper, “Injected and Leaked: Actively Inducing Side-Channel Leakage Using Electromagnetic Injection and Hardware Nonlinearity,” at USENIX Security 2026.

Traditional electromagnetic side-channel attacks rely on passively collecting radiation emitted by electronics. That’s often difficult with audio, since low-frequency signals produce weak emissions that get lost easily in background noise.

InjectEave takes a different route. An attacker transmits an electromagnetic signal toward a device at a frequency between 0 MHz and 9 MHz. The researchers did not disclose the precise settings needed for the attack.

The injected signal interacts with nonlinear parts inside the device, including amplifiers, analog-to-digital converters, power converters and switching MOSFETs. Those components can mix the injected RF signal with audio or other low-frequency activity. The device then emits a modified signal that nearby radio equipment can pick up and analyze.

The researchers used a USRP B210 software-defined radio, antennas, a Siglent SSA3075X Plus spectrum analyzer and a laptop. They also used an RF power amplifier in some tests to increase the range.

The team tested 11 commercial products. They included Sony ZX110AP wired headphones, Apple earbuds, UGreen MAX2 headphones, Philips TAH2020 headphones, HP H231R headphones and a Flyingvoice P23GW VoIP phone. The researchers also tested smart fans from Oidire and Xiaomi, as well as lamps from Jingzao and Xiaomi.

According to the paper, most tests worked at distances greater than two meters, including through walls. Device-specific ranges generally ran from one to six meters. With an RF amplifier, the researchers recovered intelligible headphone audio from up to 30 meters.

“Our new project, InjectEave, shows that RF signals can induce information leakage from everyday headphones, allowing an attacker to recover headphone audio from up to 30 meters away, including through walls,” Yan Long, an assistant professor at HKUST in Guangzhou, said in an email to The Register.

Long said the researchers confirmed the issue in devices made by Sony, HP and Philips, among others.

The team also tested scenarios involving equipment hidden in a suitcase, behind a hotel-room wall or inside office furniture. The experiments suggest the attack could be carried out outside a lab, although it still requires nearby radio equipment and knowledge of how a given device responds to the injected signal.

Headphones and phones are the most obvious targets because they may carry private conversations. But the technique could also reveal activity in a home or office. With smart lamps and fans, the team said it could capture control signals and power-use patterns that may indicate when devices are being used.

The researchers said conventional digital protections would not stop InjectEave because the leakage occurs in the analog hardware path, rather than in encrypted data or software.

“InjectEave is immune to digital defenses such as encryption, masking, and randomization, because the leakage comes from the analog path,” the researchers wrote.

They said shielding, filtering and twisted-pair wiring can reduce the amount of RF energy that reaches vulnerable components. Those measures may make the attack harder to carry out, but the researchers said they do not guarantee protection.

President Donald Trump has responded in characteristic fashion to the recent debate around AI safety — by declaring that these concerns are part of a long line of hoaxes “generated by the Radical Left Dumocrats, for purposes of destroying our Country.”

On Saturday, Trump began his musings on AI by writing on his social network Truth Social that “many people think that the words ‘Artificial Intelligence’ are inaccurate, and very ineloquent, relative to AI, or Artificial Intelligence.” So he posted a poll where followers could vote on a new name for the technology — Superior Intelligence, Extreme Intelligence, or Supreme Intelligence.  (The poll is still ongoing as of publication time) 

A couple hours later, Trump followed up with a post claiming that attempts at “the decimation, or destruction, of AI” are one of “many” Democratic hoaxes, similar to “RUSSIA, RUSSIA, RUSSIA, UKRAINE, UKRAINE, UKRAINE, Global Warming, Impeachment Hoax #1, Impeachment Hoax #2, Men in Women’s Sports, [and] Transgender for Everyone.”

Trump then insisted that he “will not stand by and let this happen,” adding that this so-called hoax “began with an attack on our Data Centers, until people realized how wealthy and prestigious they were for the Communities in which they were built.” But after, in his telling, “the crazed Data Center attack has largely failed,” critics are “going straight at AI.”

Trump did not offer any evidence to back up his claim that widespread suspicion of AI and data centers is not organic and sincere. (Both Republicans and Democrats have taken aim at data centers, and New York recently became the first state to halt permits for large projects.)

The president’s post also echoes his comments at a golf tournament last weekend, when he said he’s open to “guardrails,” but also argued, “I think you have a lot of negative forces that are bringing it up that shouldn’t be bringing it up.”

On Saturday, Trump went on to say that he will “cherish [the AI industry,] help it, and watch over it, as it grows,” but he’s also forming an AI Force, similar to the Space Force that he established in his first term.

“To that end, I will be announcing, in the near future, the AI ‘Czar’ — Only High I.Q. individuals need apply!” he added.

The president did not say what the AI Force and AI czar’s duties will entail. Venture capitalist Daivd Sacks stepped down as Trump’s AI and crypto czar earlier this year, moving on to co-chair the President’s Council of Advisors on Science and Technology.

The AI safety debate recently intensified after an AI researcher said he was quitting Anthropic over concerns that the leading AI companies “earnestly believe it could kill us all by the end of the decade” and are “gambling with our lives.” Anthropic CEO Dario Amodei subsequently released a plan to “pace the frontier,” which was seemingly endorsed by OpenAI CEO Sam Altman and SpaceX CEO Elon Musk. Critics of the AI industry have suggested that many of these concerns around the technology’s supposed existential threat are a distraction from AI’s more immediate harms.

Nvidia CEO Jensen Huang, meanwhile, called Trump on-stage during the All-In Summit (which Sacks co-hosts), agreed with the president that the AI backlash is a “hoax,” and insisted that “we’re not going to let” a slowdown happen.

Manus went back to being an independent business in September after China blocked Meta’s acquisition of the start-up.

AI start-up Manus is reportedly gearing up to double its valuation to $4bn in an upcoming fundraising round, after Meta’s acquisition of the company was forcibly unwound by Chinese authorities earlier this year.

According to Bloomberg, Manus’s fresh $500m round is set to close “soon”, after which the company could become China’s most valuable start-up in its sector. The talks are still in early stages and the outcome could change, sources told the publication.

Manus is behind a general AI agent called Manus AI that utilises multiple frontier models and operates in a complete sandbox environment to help users with multi-step tasks.

Meta acquired the Singapore-headquartered company (developed by China’s Butterfly Effect) late last December at a reported value of more than $2bn. This marked a fourfold jump from its valuation in April 2025, highlighting the perceived value of its general AI agent that took the industry by storm following its launch.

The acquisition, however, faced near-immediate scrutiny from Chinese authorities, which launched a probe shortly following the deal’s announcement.

According to national rules, the Chinese government, which is increasingly protective of its innovative technology and companies, needs to approve the export of certain technologies, including AI. Meta, at the time, said that the deal “complied fully with applicable law”.

Manus had already assimilated with Meta by the time the deal had to be unwound. Existing Manus investors, including Tencent Holdings, ZhenFund and HSG, had already received their acquisition proceeds, which needed to be reversed.

The Chinese start-up formally resumed operations as an independent business in September, months after the deal was struck down by China’s National Development and Reform Commission in April. Tech giant Tencent is now Manus’s largest external shareholder.

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.

Cybersecurity has a fundamental blind spot. We spend enormous sums protecting software while the processor underneath it blindly executes whatever instructions it receives. That has to change. For the billions of embedded systems running cars, medical devices, industrial controllers, network equipment, and critical infrastructure, security needs an independent layer that can watch processors execute instructions.

The stakes are already visible. The United States recorded 3,322 reported data breaches in 2025, a record high. Cyberattacks accounted for 80% of them. These numbers do not tell us that every defense has failed, but they do tell us something important: adding more security products has not made the underlying problem disappear.

I have spent decades building technology, and one lesson keeps returning: when a system repeatedly fails in the same place, adding another layer around it is not necessarily progress. Sometimes you have to move the security boundary.

Today, that boundary is overwhelmingly software. We deploy firewalls, endpoint protection, intrusion detection, vulnerability scanners, sandboxes, monitoring systems, and countless other tools. They are valuable. But every software defense is itself software, and software contains bugs. We are often asking vulnerable software to protect vulnerable software.

That creates another problem: noise. A network operation center can receive thousands of alerts in a day. Some are genuine threats. Others are harmless. When defenders cannot reliably distinguish between them, they eventually face the same problem doctors faced during the early days of COVID-19 testing: a test that produces too many false alarms becomes less useful, even when the underlying science is sound.

The better question is: Is the machine actually doing what it is supposed to?

A processor executes instructions at extraordinary speed, but traditionally it has no understanding of whether those instructions are legitimate. If an attacker exploits a software vulnerability, the processor may execute the malicious instructions just as obediently as the legitimate ones.

Imagine instead an independent hardware layer watching those instructions as they execute. It could enforce rules about what the software is allowed to do. A buffer overflow could still exist in the underlying code, but the processor could stop the resulting prohibited behavior before it becomes an exploit.

That is fundamentally different from asking another piece of software to detect the attack after the fact. Hardware cannot be remotely rewritten in the same way software can. It can provide a security boundary that does not depend on every line of code being perfect.

And there is a useful side effect. This kind of oversight can expose bugs before the system is deployed in the field. During normal operation, a system could identify behavior that violates its rules and give developers evidence of a vulnerability they did not know existed. Security becomes part of the software development process rather than merely an emergency response mechanism.

We desperately need that backstop because memory-safety vulnerabilities, such as buffer overflow, remain stubbornly common. According to CISA, Microsoft has reported that roughly 70% of its annually assigned CVEs are memory-safety issues, while Google has reported a similar proportion among serious Chromium security bugs. CISA also points out that these problems persist despite years of fuzzing, static analysis, sandboxing, and other testing techniques.

This matters even more as AI accelerates both sides of the fight. In February 2026, Anthropic reported that Claude Opus 4.6 had helped identify and validate more than 500 high-severity vulnerabilities in open-source software. The same capability that gives defenders unprecedented visibility can give attackers unprecedented speed. Anthropic warned that AI models are already capable of identifying novel vulnerabilities and that the traditional time available for disclosure and remediation may no longer be sufficient.

Verizon’s 2026 Data Breach Investigations Report makes the urgency even clearer. Vulnerability exploitation became the leading initial access vector, responsible for 31% of breaches in its dataset. Verizon also reported that AI is helping attackers accelerate the exploitation process.

The future can go one of two ways.

Processors become active participants in security. Embedded systems can continue operating even when software contains flaws. Developers get continuous evidence about weaknesses. Manufacturers build devices that are harder to exploit. Cars, medical equipment, industrial machinery, and connected infrastructure become more trustworthy because the security boundary sits closer to the point where code becomes action.

The alternative is darker. We keep piling software defenses onto increasingly complex software stacks while attackers use AI to find weaknesses faster than humans can patch them. The attack surface grows, alerts multiply, and the code controlling physical systems becomes harder to trust. Eventually, the gap between discovering a vulnerability and exploiting it becomes shorter than our ability to respond. The implementation of oversight not only protects your applications, but it protects your cyber defense systems, making them able to do their jobs effectively.

We should not wait for that darker future to arrive.

Memory-safe languages, better development practices, testing, patching, and conventional cybersecurity all matter. Hardware oversight does not replace them. It gives them a backstop.

So the next time you evaluate an embedded platform, a connected device, or technology that will control something in the physical world, ask a harder question than “How secure is the software?” Ask: “What is watching the processor when the software fails?”

That is where cybersecurity needs to go next

• Samsung is shifting additional DDR5 and SSD production to outside manufacturers
• The company needs internal factory space for advanced HBM production
• DDR5 module assembly remains simpler than advanced HBM packaging processes

Samsung Electronics is shifting all additional production of conventional memory modules onto outsourced partners going forward.

The company is directing capacity growth for DDR5 modules and SSDs toward external firms, reserving its in-house capacity for the more advanced HBM packaging process.

UL’s Prof Sean Fair discusses his research on bull fertility, the ‘life blood’ of an academic and how he’s using AI and machine learning.

Today (17 September) marks the final day of the National Ploughing Championships 2026, Ireland’s largest outdoor agricultural event.

The spotlight has been firmly fixed on all-things-agriculture at the event in Co Offaly this week, including the ways that science and technology are being used to advance the sector.

For example, earlier this week a number of agritech start-ups showcased their ideas at Enterprise Ireland’s Innovation Arena live pitching competition at the Ploughing Championships, with Roscommon’s VetPal taking home the top prize – the €10,000 Start-Up of the Year award.

Science and tech are more important than ever to Irish agriculture, with a number of researchers and founders alike focused on innovating and transforming various aspects of the country’s largest indigenous sector.

Prof Sean Fair is one such researcher.

Growing up on a livestock farm in Galway, Fair got a “deep hands-on start” with animals very early on.

“Caring for newborns and watching them thrive has always felt special,” he tells SiliconRepublic.com. “I never set out to become a researcher – I simply followed the questions that fascinated me and wanted to understand things more deeply.”

This curiosity eventually led Fair to a PhD in reproductive biology, where he was able to work alongside inspiring people across both research and the agriculture industry.

Now, Fair is head of the Department of Biological Sciences at University of Limerick (UL) and coordinator of the BullNet project – a Marie Curie Doctoral Training Network focused on understanding and improving bull fertility.

“I really enjoy having the academic freedom to follow my interests, and being able to progress assisted reproductive technologies in livestock allows me to be able to give back to the type of farm I grew up on,” he says.

Here, Fair talks to us about his research and the BullNet project.

Can you tell us about BullNet and the research you’re currently working on?

Global demand for meat protein is rising steadily, driven by a growing world population and a consumer preference for natural, nutrient-dense foods. Animal production systems urgently need to adopt technological efficiencies and more sustainable practices to reduce environmental pressures.

Farmers can selectively breed more genetically elite animals that grow faster, use feed more efficiently and produce fewer emissions per kilogram of meat, thereby reducing their hoofprint on the environment. Assisted reproductive technologies such as artificial insemination, sex-sorted semen and embryo technologies allow us to disseminate genes from elite animals more widely so as to enhance genetic gain for economically and socially important traits.

BullNet is a Marie Skłodowska-Curie Actions Doctoral Training Network that I coordinate, and we have 14 PhD students based across Europe working on various aspects of bull fertility.

It is focused on understanding and improving bull fertility, and comprises of a multidisciplinary and inter-sectoral research programme designed to unravel the complex underlying biology of compromised fertility of individual bulls.

We use cutting-edge basic, applied and machine-learning approaches to understand the biological regulation of sperm function and testis biology. One of the main themes of my own research is why some bulls have sperm that has normal motility and shape, as assessed under a microscope, but result in reduced pregnancy rates in the field.

We are also studying how sperm cross-communicate with both the male and female reproductive tracts, how this affects the establishment of pregnancy and how paternal epigenetic contributions are passed on to the next generation.

How is AI and machine learning transforming your research area?

We are using AI and machine learning in a systems biology approach by integrating large omics-based biological datasets (genes, proteins, metabolites) to more comprehensively understand the underlying biology of the male contribution to pregnancy establishment.

For example, we have recently shown that while only one sperm fertilises the egg, the millions of other sperm interact with the lining of the uterus and fallopian tubes and alter its environment for the benefit of the developing embryo. So it’s a team effort!

We now know that sperm need to be alive to stimulate an immune response in the uterus – and indeed, sperm from different males do this differently. Sperm from high-fertility bulls upregulate biological pathways related to embryo development and promote the establishment of pregnancy. We have also used machine learning tools to identify proteins on the sperm surface which we have validated as more reliable biomarkers of bull fertility.

In short, it brings forward biomarkers and biological pathways that we would never have thought of and allows us then to follow new avenues of research. This has implications for other species, including humans.

How can this research transform farming practices?

Being able to reliably predict the fertility outcome from a semen sample prior to its release into the field is crucial, especially for young, genetically elite bulls in their first season. Because of DNA technology, we can now identify the most elite bulls within weeks of birth. But these young bulls don’t produce functional sperm until they reach puberty at approximately nine months of age, and even then, they produce low volumes of poorer quality semen – and we don’t know their pregnancy rates in the field.

By having reliable biomarkers of semen quality, the animal breeding centres can release semen from these young bulls (both sex-sorted and conventional semen) with confidence that farmers will achieve normal pregnancy rates. On-farm fertility is a key driver of profitability on grass-based livestock farms.

‘Funding is the life blood of an academic’

What are some of the biggest challenges you face as a researcher in your field?

Constantly seeking out funding opportunities to sustain my research. Funding is the life blood of an academic and in the university sector we don’t have a core research budget. The Irish and EU funding scene is very competitive and while I am lucky to have secured a number of large-funded research projects, it is always a challenge to ensure there are not gaps in funding as we would lose staff that have built up high levels of expertise.

Are there any common misconceptions about this area of research? How would you address them?

That we are editing the genome of animals. Our focus is to work with the wider animal breeding industry and geneticists who identify the most elite animals for traits of interest and then we use assisted reproductive technologies to widely disseminate the genes from these animals.

So it’s perfectly natural and while we work with bull sperm, in reality this is only a means of disseminating DNA from the paternal line.

Is there other research in your area that you’d like to see tackled in the years ahead?

In vitro gametogenesis is a laboratory technique that turns somatic cells, such as skin or fibroblasts, into functional eggs or sperm by first reprogramming them into induced pluripotent stem cells and then guiding them through germline development.

This has been accomplished in mice, and I would like to see this being developed in livestock as it has the potential to rapidly speed up genetic progress when combined with other assisted reproductive technologies. As these technologies evolve, it is essential that we conduct well-designed experiments to monitor any off-target effects in the next generation.

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.