Users can now hold real-time voice conversations with Google’s AI-powered Search, thanks to a major new feature rollout in the Google app for Android and iOS.

Search Goes Conversational

Google this week announced the launch of Search Live with voice input, a new capability inside the Google app that allows users to engage in back-and-forth spoken conversations with its AI-powered Search tool. Rolled out first in the United States, the feature is initially available to those who have opted into the AI Mode experiment in Google Labs, the company’s testing platform for early-access features.

Hands-Free Search

The launch marks a step forward in how users interact with Search, with Google positioning the update as a more natural, hands-free way to discover and explore information while multitasking or on the move.

Use the “Live” Icon

A dedicated “Live” icon now appears within the Google app interface, allowing users to tap and speak their queries aloud. The AI responds in spoken form, and users can follow up with further questions to refine or expand the topic, thereby mirroring a more human-like back-and-forth conversation.

According to Google, Search Live “lets you talk, listen and explore in real time,” giving users the ability to access web-based information while continuing to use other apps or even switching between tasks. The tool also provides on-screen links to source material, allowing users to dig deeper into AI-generated answers.

Building on Gemini and Search Infrastructure

Search Live actually runs on a custom version of Gemini, Google’s multimodal large language model, which powers many of its generative AI tools. The Gemini model used in AI Mode has been specially adapted to support live voice input, real-time responses, and integration with Google Search’s existing ranking and quality systems.

Liza Ma, director of product management at Google Search, explained in a company blog post that the system combines “advanced voice capabilities” with the reliability of Search’s “best-in-class quality and information systems,” ensuring that responses are both conversational and trustworthy. She also confirmed the use of Google’s ‘query fan-out’ technique, which enables the system to return a more diverse and useful range of web content in response to user questions.

For example, a user might ask, “What are some tips for preventing a linen dress from wrinkling in a suitcase?” and then follow up with, “What should I do if it still wrinkles?” The AI answers audibly while presenting related links on screen. This continuity is key to what Google hopes will be a smoother, more context-aware search experience.

How and Where to Access It

At launch, Search Live with voice is available only to users in the U.S. who have joined the AI Mode experiment through Google Labs. It works on both Android and iOS via the official Google app. There is currently no timeline for a broader international rollout, though Google says it intends to expand features and availability in the coming months.

Users who have access will know because they see a new “Live” microphone icon below the search bar in the app. Once activated, they can ask a question out loud and receive a spoken response. Users can view a transcript of the interaction, continue the conversation via typing if preferred, and even revisit past queries via the AI Mode history log.

Multitask While it Works in the Background

Also, because Search Live works in the background, it enables a degree of multitasking not previously possible with voice-based search tools. For example, a user could begin a conversation in the app, switch to messaging or maps, and continue speaking to the AI without interruption.

Voice, Visuals, and What Comes Next

The introduction of voice input is actually just one part of Google’s broader plan to bring real-time multimodal capabilities into Search. For example, at Google I/O in May 2025, the company previewed future updates that will allow users to combine voice interaction with real-time visual input via their phone’s camera, building on advances made in its Project Astra research and the ongoing development of Google Lens.

Multimodal Search

This evolution represents a deeper move by Google into what’s referred to as multimodal search, whereby users can interact with AI not just through typing or talking, but by showing it what they see. In practical terms, this could include pointing the phone at a confusing diagram or damaged object, asking what it is, and getting a contextual explanation, complete with suggested web links, video tutorials or shopping sources.

It also echoes the direction competitors are taking. For example, OpenAI’s ChatGPT has recently introduced voice interaction capabilities in its mobile apps, and Perplexity AI has gained traction for its own real-time web search and voice tools. Google’s response, with Search Live, is both a defensive and strategic step to stay ahead in what is quickly becoming a crowded, AI-first search market.

A New Frontier for Business and Advertisers?

For business users, the implications of voice-first search are far-reaching. For example, in sectors such as logistics, retail, and field service, the ability to conduct voice-based queries while driving or working could prove invaluable. Search Live also introduces potential benefits for productivity, especially for knowledge workers trying to conduct research or fact-checks while multitasking between devices or applications.

It may also signal a new phase for Google’s advertising ecosystem, although details remain unclear. As Search becomes more conversational and voice-led, traditional search result ads, particularly those dependent on text input and visual scanning, may need to evolve. It’s not yet known how, or if, Search Live results will incorporate sponsored content.

The visual links shown alongside voice answers could potentially become prime real estate for future advertising formats. However, Google has so far remained quiet on how monetisation will work within AI Mode. With more users consuming answers audibly and potentially clicking fewer links, publishers and advertisers will be watching closely.

Challenges

Despite the promise, it should be noted that there are several challenges ahead. For example, accuracy and reliability remain key concerns for AI-generated search responses. While Google stresses its Gemini-based AI uses the same quality controls as regular Search, AI hallucinations (where systems confidently give false or misleading answers) are still a known risk in generative models.

The opt-in nature of the feature also limits immediate user exposure and feedback. By placing Search Live behind the AI Mode experimental wall, Google is clearly seeking to manage rollout cautiously but this also means that the majority of users globally still can’t access or evaluate it.

There are also privacy and data security implications, particularly with voice-based input and persistent conversation histories. Google maintains that users can view, manage or delete their AI Mode interactions, but questions remain over how voice data is processed, stored, or used to train models.

One other aspect critics may point to is the increasing opacity of sources in AI answers. For example, while Google includes clickable links alongside Search Live responses, these can sometimes appear secondary to the spoken reply, which may not fully represent the nuance or breadth of available information. Ensuring transparency and balance in summarised answers will be crucial to maintaining trust, especially as Search Live expands into more domains.

What Does This Mean For Your Business?

The introduction of Search Live could be seen as the next step in its natural progression towards Google’s long-term vision for AI-powered search. By blending real-time voice interaction with the depth of web content, Google is essentially positioning itself not just as a search engine but as a more intuitive, responsive assistant capable of handling everyday queries in more dynamic, human-like ways. However, the fact that it’s limited to U.S.-based testers in Labs signals Google’s awareness of the stakes involved. It is not just testing technology but testing trust, usability and commercial viability all at once.

For UK businesses, this could open up important new opportunities once rolled out more widely. Voice-driven interaction with AI may reduce the need for screen time in roles where hands-free efficiency matters, i.e. from trades and transport to healthcare and hospitality. It could also help knowledge workers process information faster while juggling tasks, potentially enhancing productivity and reducing friction in routine research or client support work. There are potential implications for business intelligence and even internal training, particularly once real-time camera input is layered in. But these benefits will only be realised if the underlying AI delivers reliable and verifiable responses at scale.

Advertisers and content publishers are likely to be more cautious. With fewer visual interactions, the conventional search engine results page model may weaken. If users hear an answer but don’t tap the links shown, that affects traffic and engagement metrics. This will raise fresh questions about how brands position themselves within voice-first search and whether new advertising formats will emerge within AI Mode or remain separated. Also, the monetisation path here is still not altogether clear and, as Google experiments with form, it may need to reassure partners that function won’t entirely override visibility.

Meanwhile, Google’s competitors such as OpenAI and Perplexity AI will, no doubt, be watching closely. Each is racing to define the next evolution of everyday search, combining voice, visuals and real-time reasoning. Google still has the infrastructure advantage, but the race is no longer just about data—it’s about usability, privacy, and user confidence. In that context, Search Live’s success may depend as much on how it is governed and explained as how well it works technically.

Whether Search Live becomes the new normal or remains a feature for power users will likely depend on the clarity of its responses, the transparency of its sources, and the ease with which users (especially businesses) can trust it as a tool rather than a black box. What is clear already is that Google is laying groundwork for a future where the way we search is no longer typed, but spoken, shown and responded to in real time. Once mainstream, that could fundamentally change how we interact with the web.

In this Tech Insight, we look at why clicking the ‘unsubscribe’ link in an email might not be as safe as it seems, and how cybercriminals are using this tactic to profile victims, deploy phishing attacks, and gather intelligence for future scams.

Why the Unsubscribe Link Isn’t Always Safe

The warning comes from TK Keanini, Chief Technology Officer at cybersecurity firm DNSFilter. Speaking recently to The Wall Street Journal, Keanini explained that unsubscribe links embedded in spam emails are increasingly being used by cybercriminals as a means of identifying active users and directing them to malicious websites.

Not Just Theoretical

The risks are not just theoretical. For example, DNSFilter estimates that roughly one in every 644 clicks on an unsubscribe link leads to a harmful destination. That may sound like a small percentage, but across the billions of marketing emails sent each day, the number of victims quickly adds up.

Unlike legitimate unsubscribe tools offered by trusted senders, these deceptive links don’t remove you from a list. Instead, they exploit your trust—by either redirecting you to phishing pages designed to steal your personal information, or by quietly logging your interaction to flag your email address as a ‘live’ target for further attacks.

What Makes These Links So Dangerous?

Keanini warns that while many spam emails are caught by filters, some still slip through. Also, when users click the unsubscribe link at the bottom (thinking they’re taking control of their inbox) they’re often doing the exact opposite.

“There’s a big difference between the unsubscribe function embedded by your email client and the one coded into the email itself,” Keanini explained. “The latter can send you out of the protected environment of your email platform and onto the open web, where you’re far more vulnerable.”

At best, this action notifies scammers that your address is actively monitored. At worst, it takes you to a spoofed landing page where you might be asked to enter your email address or login credentials under false pretences. Some pages can even exploit vulnerabilities in your browser to initiate malware downloads or install tracking scripts.

Security analysts have also warned that even a single click can help attackers build up a profile on a target. Over time, this can lead to more personalised phishing emails, fake login pages, or even ransomware attacks disguised as legitimate follow-ups.

Better Ways to Unsubscribe Safely

Fortunately, there are safer ways to manage unwanted emails. Most modern email clients, including Gmail, Outlook, Apple Mail and others, use a function known as list-unsubscribe headers. These headers are recognised by the email platform and often display a safe, in-built unsubscribe button near the top of the message, such as Gmail’s “Unsubscribe” link next to the sender’s name, Apple Mail’s grey “Unsubscribe” button below the subject, or Outlook’s banner option above the message content.

Since list-unsubscribe headers are rendered by the email provider itself (not the email sender) they don’t carry the same risks and, therefore, act as a kind of trusted bridge between you and the sender’s database (if that database exists at all).

Just Mark it as Spam or Block the Sender

If no list-unsubscribe option is present, experts recommend marking the message as spam, blocking the sender, or setting up an automated filter. In some cases, you can even block the sender’s IP address if they persist in using different email accounts.

Use Disposable Email Addresses

Another good practice is using email aliasing or disposable addresses. Gmail, for example, supports ‘plus addressing’, which lets users sign up to services using addresses like yourname+shopping@gmail.com. If that alias starts receiving spam, you can simply filter or delete it without affecting your main account.

Apple’s ‘Hide My Email’ feature offers a similar layer of privacy, creating unique, random addresses that forward to your inbox. This helps mask your real address from third parties and allows you to shut down addresses that become compromised.

Businesses and Marketing Teams

While this development raises new concerns for individuals, it also carries implications for legitimate businesses that rely on email marketing. For example, if users start to fear unsubscribe links, they may avoid interacting with even trusted messages, making it harder for businesses to stay compliant with laws like the UK’s Privacy and Electronic Communications Regulations (PECR) or GDPR.

Under these laws, all commercial emails must include a clear and effective opt-out mechanism. But if users don’t trust that mechanism, businesses may find themselves facing both technical and reputational risks.

Email marketers are now being encouraged to make use of trusted unsubscribe headers recognised by major email clients, rather than relying solely on HTML links in the message body. Tools like Mailchimp, HubSpot, and Campaign Monitor already support these built-in mechanisms, which reduce the need for external web redirects and improve user trust.

Really, therefore, transparency is key. Making sure that unsubscribe options are clear, legitimate, and functional will go a long way in protecting both customers and brands from reputational fallout or false positives in spam filters.

Business Users at Higher Risk

For business users, especially those using personal emails for professional tasks, the risks of phishing and malware attacks are actually significantly higher. For example, a successful scam could lead to leaked client data, ransomware disruption, or credential theft that compromises cloud-based systems and internal communications.

Businesses should, therefore, ensure staff are trained not to click unsubscribe links in suspicious or unexpected emails, even if they appear to be from reputable sources. Phishing simulations and email security briefings can help reinforce this behaviour.

Keanini points out that malicious unsubscribe links are unlikely to be the attacker’s only tool. “Often, it’s part of a larger campaign,” he noted. “They’re looking for a response—any sign that there’s a human on the other side. Once they get that, they plan their next move.”

Safer Email Solutions for Businesses

Organisations looking to harden their defences should perhaps consider adopting enterprise-grade email protection tools that go beyond simple spam filtering. For example, providers like Proofpoint, Mimecast, and Barracuda (there are others) offer advanced threat protection that scans links in real-time, blocks phishing attempts, and provides safe-click technology.

Microsoft 365 and Google Workspace users can also leverage built-in protections such as Safe Links, quarantine reviews, and anti-spoofing measures to prevent dangerous emails from ever reaching end users.

Zero-trust email platforms are gaining traction as well. Tools like Proton Mail for Business and Tutanota offer end-to-end encryption, IP address masking, and strict sender verification, all designed to limit the exposure of user identities and block malicious redirections.

Cybersecurity Best Practices for Email

In addition to technical tools, businesses should encourage staff to follow core email hygiene principles, such as:

– Never click links in unsolicited or unfamiliar emails.

– Hover over links to preview the actual destination URL.

– Use multi-factor authentication (MFA) on all email accounts.

– Regularly update antivirus and anti-malware software.

– Report suspicious emails to the IT or security team for review.

– Conduct quarterly training on evolving phishing tactics.

By implementing a layered approach, combining user awareness, secure infrastructure, and smart email practices, organisations can drastically reduce the likelihood of falling victim to these increasingly sophisticated scams.

What Does This Mean For Your Business?

What this ultimately shows is that something as familiar as clicking an unsubscribe link can carry far more risk than most users realise. While many will continue to treat email as a low-risk tool, the reality is that attackers are exploiting habits formed over years of legitimate marketing interactions to identify targets and launch broader attacks. This makes the unsubscribe link not just a nuisance, but a potential entry point into much more serious compromise.

For UK businesses, this means rethinking not only how they engage with their own inboxes but also how they structure outbound communications. Any marketing email must now earn trust, not just attention. That means using secure, standards-based unsubscribe methods and making it absolutely clear to recipients that their data is being handled properly. Businesses that fail to do this may find their messages ignored, filtered or marked as suspicious, with reputational consequences that go far beyond email.

At the same time, internal safeguards matter more than ever. Many business users still use personal inboxes for work tasks or operate without layered protections in place. With phishing emails now frequently designed to look like marketing communications, the boundary between personal and professional threat surfaces has blurred. IT teams must assume that not all employees will know the difference between a safe unsubscribe link and a dangerous one, and must build protections around that assumption.

The wider lesson here is that, whether individuals, businesses, or email service providers, even routine digital interactions need to be scrutinised in today’s threat landscape. Protecting users now means going beyond spam filters and encouraging safer behaviour at every level, from the tools people use to the training they receive. It seems that the unsubscribe button, once a symbol of user control, now serves as a reminder that even good habits can be weaponised if they’re not re-evaluated through a security lens.

A Berlin-based health tech company has received official European approval for its at-home saliva fertility tracker to be marketed as a certified contraceptive device, a first of its kind.

From Fertility Tracker to Certified Contraceptive

The product in question, called the Minilab, was developed by Inne, a women’s health startup founded by entrepreneur Eirini Rapti. Until now, the Minilab has been used primarily as a tool for tracking fertility and menstrual cycles, marketed to women trying to conceive. However, following a clinical trial and regulatory review, the device has now been certified as a medical contraceptive across Europe.

BSI

The approval came from the British Standards Institution (BSI), one of Europe’s major medical device certifiers, and paves the way for the Minilab’s roll-out across the EU, starting in Germany and Austria, with UK availability expected later this year.

An Effective Hormone-Free Alternative

What sets the Minilab apart is that it is non-invasive, hormone-free and digital-first. For example, instead of using oestrogen or progestin to suppress ovulation, or inserting physical devices like IUDs, the Minilab works by measuring a key reproductive hormone (i.e. progesterone) in saliva, thereby offering a precise read of a woman’s cycle in real time.

100% Effective In Study

A one-year study involving 300 women across 1,467 cycles found the method to be 100 per cent effective with perfect use, meaning no unprotected sex during the identified fertile window, and 92 per cent effective under typical use conditions. That puts it in line with the combined contraceptive pill (99 per cent perfect use, 93 per cent typical) and more effective than condoms (98 per cent perfect, 87 per cent typical), according to NHS data.

Caveat

However, it’s worth noting that the study was observational, relatively small, and not peer-reviewed, meaning further validation will be needed over time to bolster long-term trust.

The Science Behind the Saliva

In terms of the science behind it, the Minilab uses a lateral flow assay (a test format similar to COVID-19 and pregnancy tests) in combination with antibodies that react to progesterone. Each morning, users deposit a small amount of saliva on a test strip, which is then inserted into a pocket-sized electronic reader. Over 10 minutes, the device photographs and analyses how hormone particles move along the strip.

Companion App

The data is processed using image recognition and biochemistry algorithms, then synced with a companion app. The app displays hormone levels, indicates the user’s current fertility status, and highlights “high probability” days for pregnancy risk.

According to Rapti, this offers a more accurate and personalised experience than alternatives such as temperature-based methods (used by competitors like Natural Cycles) or calendar-based period tracking apps.

“Temperature can be affected by illness, disrupted sleep, or alcohol,” she explained (in an interview with Euronews Health). “Saliva gives you direct insight into hormonal changes — it’s biological data, not pattern recognition.”

Who Can Use It And Who Can’t?

While Inne is pitching the Minilab as a more comfortable and flexible method, it isn’t suitable for everyone. The company only recommends it for women who:

– Are over 18.

– Are not currently pregnant or breastfeeding.

– Have regular menstrual cycles (22 to 35 days).

– Are not taking hormonal contraceptives or other hormone-altering treatments.

– Have completed at least two full menstrual cycles after stopping hormonal contraception, pregnancy, or breastfeeding.

Flexible Test Window

The test window is flexible, allowing women to take it within a four-hour period each day, and everyday influences like minor illness, sleep quality, or alcohol consumption don’t affect the saliva readings, according to Inne.

Subscription-Only (For Now)

For now, the Minilab is available on a subscription basis, starting at €24 per month when paid upfront for two years. The device is already covered by Germany’s largest public health insurer, and the company is in discussions to expand insurance coverage to more markets.

A Sign of the Times in FemTech

Inne’s move into contraception could be said to reflect a broader trend in femtech, i.e., the intersection of digital health and women’s wellbeing, where user demand for hormone-free alternatives is growing. Apps like Natural Cycles and wearables like Daysy have gained traction, but most still rely on indirect indicators like temperature or physical symptoms.

Several other companies are also innovating in this space. Natural Cycles, based in Sweden, remains the most established, offering an app-based contraceptive approved by EU regulators and the US FDA, which relies on temperature readings and cycle tracking. US-based Oova uses urine hormone testing paired with an AI-driven app to support both conception and cycle tracking, while Mira offers a home hormone analyser that measures luteinising hormone (LH) and oestrogen metabolites using urine samples. Though not currently certified as contraceptives, these tools reflect a growing shift toward bio-data driven reproductive management.

Minilab’s focus on hormone measurement aims to make fertility awareness methods, which have long been viewed as unreliable, scientifically robust and medically certified. Also, unlike blood hormone tests, saliva offers a non-invasive, low-cost, and scalable home testing method.

“We are excited and proud to offer women a modern, safe, and hormone-free method that enables them to take charge of all aspects of their fertility,” said Eirini Rapti, Inne’s founder and CEO, in a statement on the company’s website. “This includes conception support, cycle tracking, and now — contraception.”

Not Without Its Limits

Despite the innovation, Inne’s Minilab is not a silver bullet. For example, the clinical study behind its certification has not been peer-reviewed and included only around 300 participants, which is a far cry from the multi-year, tens-of-thousands scale typically used to validate pharmaceutical contraceptives. There was also no control group to rule out behavioural bias.

It’s also worth noting here that like other fertility awareness methods, the device requires consistent, disciplined use, especially around the fertile window. While “perfect use” returned zero unintended pregnancies in the study, real-life adherence may prove more challenging.

As the Pearl Index for typical use (7.98) indicates, up to 8 in 100 women may still become pregnant annually which is something that regulators and healthcare providers will need to communicate clearly as adoption expands.

A Regulatory and Approval Milestone

For Inne, regulatory approval marks a major milestone in a mission to reshape reproductive health. It opens doors for reimbursement discussions, partnerships with public health bodies, and a foothold in the lucrative digital contraception market.

However, it also sets a precedent for other non-invasive, hormone-free technologies to follow. Natural Cycles, which became the first certified app-based contraceptive in 2017, paved the way for digital fertility solutions. Inne now adds a saliva-based dimension that could push the sector further toward precision diagnostics and preventative care.

Plans To Track Other Hormones

It seems that plans are already underway to extend the Minilab platform to track additional hormones like cortisol and testosterone, which could unlock applications in stress management, athletic performance, and broader hormone health.

A New Category?

As women increasingly demand options that are accurate, private, non-invasive, and side-effect free, regulators and insurers may be forced to rethink how they categorise and fund digital tools that sit between medicine, lifestyle, and health tech.

What Does This Mean For Your Business?

The regulatory approval of Inne’s Minilab not only introduces a new contraceptive option but also opens up a fresh category of health technology, i.e. one that blends medical-grade diagnostics with consumer-friendly usability. For women seeking hormone-free, side-effect-free alternatives, the Minilab represents a significant step forward in personalised reproductive care. Its arrival could prompt greater scrutiny of conventional methods and accelerate demand for more individualised solutions that place data and agency directly in users’ hands.

For healthcare professionals and regulators, the challenge now is how to support innovation without lowering the bar for safety and efficacy. The Minilab may be CE-certified, but the supporting evidence is still limited in size and scope. That means clear communication with users will be vital. Transparency around what “perfect use” actually requires, and how the device should and should not be used, will need to be prioritised across clinical guidance, marketing, and insurance frameworks.

UK businesses working in health tech, diagnostics, and femtech could see this as a signal to act. With the UK still aligning with many EU medical device regulations post-Brexit, and the BSI playing a dual role in both jurisdictions, there is a clear route to market for British innovators who can offer similarly precise, non-invasive, and hormone-free products. Private healthcare providers and employers may also begin to consider partnerships with platforms like Inne as part of wider wellbeing packages, particularly for staff cohorts seeking natural and flexible contraception options.

It seems that the wider digital contraception space is now under pressure to raise its own standards. While apps based on temperature or calendar data were once seen as disruptive, they now face competition from tools offering real-time biological insight. Saliva testing brings a new level of scientific grounding that could redefine what qualifies as reliable cycle tracking, especially as consumer expectations rise.

Inne may have been the first to gain certification for a saliva-based contraceptive, but it is unlikely to be the last. What happens next will depend on continued evidence, smart regulation, and how successfully these tools can prove their value not just in lab settings, but in the unpredictable realities of everyday life.

A new study from the Massachusetts Institute of Technology (MIT) has found that using generative AI tools like ChatGPT can significantly reduce brain activity during writing tasks, and may impair users’ ability to retain information and think critically.

Who And Why?

The research was led by Dr Nataliya Kosmyna, a scientist at MIT’s Media Lab, and was motivated by growing concerns about the cognitive impact of widespread reliance on AI tools in education and professional life. For example, the team said they wanted to understand how large language models (LLMs) like OpenAI’s GPT-4o affect human cognition when people use them to perform tasks traditionally done unaided, such as writing short essays.

While AI is often praised for increasing productivity, Kosmyna and her team wanted to test whether that convenience actually comes at a cognitive cost. “This is not about calling AI bad,” she told The Register. “But it’s important to understand the trade-offs, especially in learning contexts.”

How the Study Was Carried Out

The peer-reviewed preprint, titled Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task, was based on four essay-writing sessions involving 54 Boston-area university students. The study divided participants into three groups:

– One group wrote essays entirely unaided (the “Brain-only” group).

– Another used traditional search engines for research.

– The third relied on OpenAI’s GPT-4o for assistance.

In each session, students had 20 minutes to write on a given topic. Participants wore electroencephalogram (EEG) headsets that monitored neural activity throughout. In the fourth and final session, some students were switched to the opposite group, so LLM users had to write unaided, and unaided writers were allowed to use AI.

What the Brain Scans Showed

The results, the researchers said, were “striking.” EEG readings revealed that the Brain-only group consistently showed the highest levels of brain activity. These participants demonstrated stronger and more distributed neural connectivity patterns associated with cognitive load, specifically in regions involved in attention regulation, memory encoding, and semantic processing.

In contrast, those using AI tools showed up to 55 per cent lower brain connectivity, with reduced activity across all EEG frequency bands. Even search engine users showed a notable drop of 34–48 per cent, depending on the task.

The metric used to assess this, known as the Dynamic Directed Transfer Function (dDTF), reflects how different parts of the brain communicate during complex cognitive tasks. The LLM group, the study noted, “elicited the weakest overall coupling,” suggesting lower engagement of executive functions and internal synthesis.

Why This Matters for Learning and Memory

The implications for learning were clear. For example, participants in the Brain-only group scored higher on factual recall tests and reported greater “ownership” of their work, meaning they were more likely to remember and understand what they had written. By contrast, LLM users struggled to recall even basic content from their own essays.

As Kosmyna explained, the use of AI may result in what the study terms “cognitive debt”, i.e., where offloading thinking to an AI tool results in weaker internal encoding of information. “The AI is doing the heavy lifting,” she said, “but that means your brain isn’t.”

Participants who switched from AI to unaided writing in the final session performed particularly poorly. When stripped of LLM support, this group exhibited what researchers called “under-engagement,” with marked reductions in both alpha and beta brainwave connectivity. In other words, having relied on AI too heavily, they struggled to switch back to independent thought.

Are There Any Benefits to Using AI for Learning?

Interestingly, participants who moved from Brain-only writing to LLM-assisted writing performed relatively well. Their brain activity remained high, and they produced more cohesive content, likely because they had already internalised the structure and subject matter.

This, the researchers say, supports a “delayed AI integration model” where learners are encouraged to first engage deeply with material unaided, and only later use AI tools to support or extend their thinking.

“Taken together, these findings support an educational model that delays AI integration until learners have engaged in sufficient self-driven cognitive effort,” the MIT team wrote in their report.

What This Means for AI Companies

For AI developers and edtech firms, the findings appear to present a potential credibility challenge. For example, while tools like ChatGPT can help streamline tasks and improve output, they may also short-circuit the very mental processes that build long-term knowledge and critical reasoning.

That said, the study does not suggest AI tools are inherently harmful, only that their role in education and professional development needs to be more carefully designed.

With growing interest from schools, universities and corporate training providers in AI-powered learning, the MIT findings may prompt calls for clearer usage guidelines and pedagogical frameworks that preserve the value of cognitive effort.

Business and Professional Users

For business users, especially those in knowledge-based sectors, the study raises pressing questions about productivity versus proficiency. Using ChatGPT to draft reports or generate marketing copy may save time, but at what cost to understanding, originality or professional development?

The study’s findings also suggest that while AI may improve speed and surface-level output, it could impair memory retention and reduce long-term mastery of content. This could have downstream effects in areas like strategic thinking, problem-solving, and even leadership development, where depth of understanding is key.

As more firms integrate LLMs into workflows, there’s likely to be a growing need for “cognitive balance” where AI tools are used not as crutches, but as scaffolds, supporting rather than replacing human effort.

Challenges and Limitations

It should be noted that, although the MIT study has gained widespread attention, it is still a preprint and not yet formally peer-reviewed. That said, its methodology is sound, and the EEG-based approach gives it added credibility compared to survey-based or observational studies.

One challenge is that the research focused on a narrow task, i.e. essay writing in an academic context. Therefore, it remains to be seen whether similar cognitive reductions occur across other task types such as coding, design, or strategic planning. Further research is also needed to examine how age, profession, or digital literacy might influence outcomes.

Also, some critics caution against drawing sweeping conclusions. As Professor Rose Luckin of University College London noted in related commentary, “The goal shouldn’t be to avoid AI, but to use it wisely. The key is metacognition—knowing when to trust it, when to question it, and how to engage critically.”

What Does This Mean For Your Business?

The findings from MIT present a clear warning about how AI tools are integrated into everyday work and learning. When individuals rely too heavily on chatbots like ChatGPT to generate content, their brains show measurable signs of disengagement. This isn’t a theoretical concern but is backed by brain scans, reduced recall, and weaker understanding of material, even when judged by both human teachers and AI systems.

For UK businesses, especially those investing in AI-driven productivity tools, the message is that efficiency gains are valuable, but they should not come at the expense of long-term knowledge development or independent thinking. Over time, a workforce that regularly outsources complex tasks to AI may become less capable of synthesising information, solving novel problems, or retaining strategic insights. In sectors that depend on deep subject knowledge, such as finance, law, consultancy, and research, this cognitive drift could affect decision quality and competitive advantage.

There are also implications for HR and training leaders. For example, if employees are encouraged to use LLMs without clear guidance, the result could be surface-level output with limited learning taking place. Onboarding, professional development, and internal knowledge-sharing may all suffer if too much cognitive effort is handed off to AI. Equally, if properly managed, businesses can use these findings to strike a better balance. Encouraging employees to first think through a problem themselves before turning to AI for comparison or refinement may protect both quality and learning outcomes.

For educators, tech developers, and policymakers, the study adds to a growing body of evidence that AI tools should not be treated as neutral helpers. Their design and usage shape behaviour and cognition, sometimes in ways that reduce intellectual ownership or capability. It highlights the need for guardrails, i.e. educational strategies that build critical thinking before AI access, interface features that encourage user reflection, and corporate policies that treat AI as a partner rather than a substitute.

The bigger challenge may be cultural. As AI becomes more ubiquitous and its outputs more polished, the temptation to default to it will grow. This study shows that without conscious effort to preserve our own cognitive involvement, we risk weakening the very faculties that made these tools possible in the first place. For any organisation that values deep expertise, the long-term costs of unchecked reliance on AI may outweigh its short-term gains.

A new AI-powered legal startup backed by Sequoia Capital is rewriting how contracts are reviewed, by building a law firm around the software itself.

Lawyers Using AI To Deliver Services To Clients

Most legal tech startups position themselves as tools for traditional firms to use. It seems, however, that Crosby has taken a radically different approach. Rather than offering AI software to outside lawyers, Crosby has built its own law firm, staffed with lawyers who use its proprietary legal AI systems to deliver services directly to clients.

Launched in early 2025 and already out of stealth with a $5.8 million seed round led by Sequoia Capital, Crosby is a hybrid legal provider combining full legal oversight with rapid AI-powered contract processing. In doing so, it positions itself not just as a legal technology provider, but as a legal services business with a completely different operating model.

“Our goal was never to just automate tasks for law firms,” said Ryan Daniels, Crosby’s co-founder and CEO, in a launch statement. “To really fix how slow legal work is, we had to control the entire process—so we became the law firm ourselves.”

Built for Speed, Designed for Growth

The problem Crosby says it set out to solve is a familiar one in fast-moving industries, i.e. contract delays.

Daniels, who previously served as general counsel for several startups and worked at elite tech law firm Cooley, experienced the issue first-hand. “Most of the time I was spending on legal was for our contracts, sales agreements, MSAs,” he said. “It was the reason we weren’t growing as fast as we wanted to.”

Agentic

Crosby’s solution is what it calls an “agentic” law firm using a “hybrid” model where every contract is reviewed by proprietary AI agents, then verified by experienced lawyers. This human-in-the-loop setup enables clients to get back a reviewed contract within three hours, with many returned in under 60 minutes. Daniels claims the company’s fastest reviews take just minutes.

High-Volume, Sales-Related Agreements

The startup focuses on high-volume, sales-related agreements such as master service agreements (MSAs), data processing agreements (DPAs), and non-disclosure agreements (NDAs). This is because these are the kind of documents that tend to clog deal pipelines for sales teams in growing firms. Crosby’s promise is, therefore, to get contracts reviewed quickly and accurately, so deals close faster.

Who’s Behind Crosby?

Crosby was founded by Daniels and John Sarihan, who serves as CTO. Sarihan previously worked at Ramp, a fintech unicorn, and brought with him engineering talent from companies like Meta, Google, and Vanta. Daniels, a second-generation lawyer whose parents are both law professors, focused on building the legal team, which includes alumni from Harvard, Stanford, and Columbia Law.

The company is headquartered in New York and operates as both a legal technology company and a law firm. Formally, Crosby Legal, Inc. provides the technology, while Crosby Legal PLLC is the law firm offering legal services.

Their Sequoia-led funding round also included participation from Bain Capital Ventures and notable angel investors such as Ramp co-founders Eric Glyman and Karim Atiyeh, Instacart co-founder Max Mullen, Opendoor’s Eric Wu, and Flatiron Health founders Zach Weinberg and Gil Shklarski.

Josephine Chen from Sequoia, who previously backed AI procurement startup Venue (later acquired by Ramp), led the deal. “Legal is a bull’s-eye case for the use of LLMs,” she said. “Contract negotiations can be a real bottleneck for growth.”

AI Meets Legal Expertise

Crosby’s approach blends the rapid processing power of AI with legal precision. For example, clients can send documents or queries via Slack, email, or through a CRM trigger. The system’s legal AI agents, trained on thousands of contracts and guided by firm-developed benchmarks, then analyse the documents, make suggestions, and insert relevant market terms.

Lawyers then step in to review, interpret tricky clauses, and validate any automated changes. The final contract is returned to the client with a fixed price tag (no hourly billing), and no redline confusion (no back-and-forth edits on contracts). For example, “AI never sleeps,” says the firm’s website. “Crosby never gets backlogged.”

Learns About Clients’ Businesses

Crosby’s AI systems are also designed to learn each client’s business over time. This includes storing preferences, preferred clause variations, and common fallback terms. The company claims its software can answer routine contract questions without client input once it’s sufficiently trained.

Targeting Startups That Need to Move Fast

So far, Crosby appears to have aimed its services at venture-backed startups, particularly those with aggressive go-to-market (GTM) strategies. Early clients include Cursor, UnifyGTM, and Clay, all startups known for rapid growth and high sales velocity.

By focusing on sales contracts and offering legal reviews as fast as the sales cycle itself, Crosby is positioning itself as a growth enabler rather than just a legal resource. GTM teams reportedly call it a “secret weapon” for getting contracts over the line.

Crosby’s upfront pricing is also designed to appeal to startups used to controlling costs. For example, clients pay per document, not per hour, which is a sharp contrast with traditional legal billing models.

Why This Matters for the Legal Industry

Crosby’s emergence poses direct questions to the traditional legal services model. For example, most law firms are structured around bespoke work, hourly billing, and long timelines. By contrast, Crosby is productising contract review, treating it as a repeatable, scalable service.

Not The First Legal Firm To Apply AI

It’s worth noting here that Crosby is not the first to apply AI to legal work. For example, companies like Harvey (which recently raised $80m), Ironclad, and Spellbook are building AI tools to support lawyers. However, Crosby is unusual in that it delivers end-to-end legal service directly to clients, with its own regulated legal team and a law firm structure.

This allows Crosby to sidestep law firm conservatism and scale more like a tech startup. “We didn’t want to wait for firms to catch up,” Daniels said. “We wanted to prove it could be done.”

Potential Risks and Criticisms

Crosby’s model is not without its critics. Legal work carries significant liability, and while its lawyers remain in the loop, the firm must prove that its AI systems are reliable, auditable, and ethically sound. The startup says all outputs are lawyer-reviewed, but how clients interpret that balance between machine and human may vary.

There’s also the regulatory question. In most US states, legal services must be delivered by licensed professionals. Crosby’s dual-entity structure is designed to comply with those rules, but regulatory scrutiny may increase as it scales.

UK firms will also need to watch this space closely. For example, while firms like Allen & Overy and Mischon de Reya are experimenting with AI copilots, none have yet adopted a Crosby-style hybrid structure. If Crosby proves successful in the US, it may set a precedent for how AI-led legal services could evolve in other jurisdictions.

Are There Any Competitors Doing the Same?

There are firms inching toward similar models. Atrium (now defunct) once tried to integrate software with legal service delivery, though without the speed or AI emphasis Crosby offers. More recently, firms like Lawtrades and Axiom Legal blend tech-enabled platforms with lawyer marketplaces, but again, they stop short of Crosby’s embedded, AI-first, regulated law firm model.

In the UK, companies like Luminance and Robin AI provide AI tools to assist legal teams but do not operate as regulated firms themselves. Crosby’s core differentiator is that it is both the software company and the law firm, acting as one unified entity with aligned incentives to deliver speed and accuracy at scale.

What Does This Mean For Your Business?

For law firms, Crosby represents a direct challenge to long-established business models built around hourly billing and drawn-out negotiations. Its hybrid setup shows that legal services can be fast, fixed-price, and scalable, without sacrificing human oversight. If the model proves durable, it could force traditional firms to rethink both their pricing structures and the level of tech integration in their workflows.

For UK businesses, the implications could be equally significant. If models like Crosby’s reach the UK market, startups and scaleups would most likely be able to close deals more quickly, reduce legal overheads, and compete more effectively. The demand for faster legal execution is not limited to Silicon Valley. UK firms under pressure to accelerate growth and reduce friction in sales cycles may soon expect legal services to move at the same pace as their CRM or procurement systems. Legal firms serving these clients will need to respond accordingly.

Regulators and legal educators may also come under pressure to modernise. Crosby’s model blurs the line between legal practitioner and product developer. That raises questions not just about compliance, but also about professional training, ethical oversight, and the future identity of the legal profession. As AI models evolve, the challenge will be to strike a balance between innovation and accountability.

The legal industry has long been insulated from the kind of disruption seen in finance or logistics. Crosby’s approach suggests that insulation may be starting to wear thin. Whether it becomes the norm or remains an outlier, it has already expanded the conversation around what legal services can look like, and who is best placed to deliver them.

A new platform called Psst has launched to help tech and government workers report wrongdoing anonymously and securely.

Users submit encrypted text-only reports into a “digital safe” at www.psst.org. These stay locked unless others report similar issues, helping protect identities and reveal patterns of misconduct.

Only Psst’s legal team can access matched reports, ensuring legal privilege and shielding whistleblowers from retaliation.

The tool avoids uploads to reduce traceability and plans to automate matching using secure hardware enclaves.

Psst arrives amid growing concern that insiders fear speaking out on security, safety and ethics—especially in fast-moving tech sectors.

Businesses should review their own reporting systems and ensure staff can raise concerns safely and confidentially.