Tag: Google

Categories
Technology

Google unveils Nano Banana 2.1 for cheaper AI images

Google has introduced Nano Banana 2.1, a new artificial intelligence image-generation and editing model designed to make high-quality visual creation faster, more consistent and significantly cheaper.

The model, launched on October 6, builds on Google’s Nano Banana 2 and focuses on a challenge that is becoming increasingly important as AI-generated images move from experimentation into everyday business use: producing better visuals without pushing up costs.

One of the biggest changes is the pricing. Through the Gemini API, Nano Banana 2.1 is available at $0.0336 for a 1K image, $0.0504 for a 2K image and $0.0756 for a 4K image. The pricing is roughly half that of the previous Nano Banana 2 model, potentially making large-scale image generation more affordable for developers and businesses.

The lower cost could be particularly useful for companies that generate thousands of images for online stores, advertising campaigns, publishing, gaming and social media. Instead of treating AI image generation as an occasional creative tool, businesses can increasingly use it as part of regular production workflows.

Google has also focused heavily on consistency. One of the common problems with AI-generated images is that a person, product or object can change slightly every time a new image is created. Nano Banana 2.1 is designed to improve that consistency across multiple editing and generation steps.

The model can work with as many as 14 reference images, allowing users to provide more visual information about the people, products or objects they want to include. It can support up to four characters and 10 objects in a scene, giving creators greater control over complex compositions.

Text inside AI-generated images has also received attention. Earlier image models often struggled with spelling, typography, signs, labels and detailed layouts. Nano Banana 2.1 is designed to produce more accurate text and better-organised infographic-style visuals.

That improvement could make the technology more practical for marketing material, product presentations, educational graphics and social media content, where incorrect text can quickly make an otherwise impressive image unusable.

The model supports 1K, 2K and 4K image generation, giving users the option to choose quality based on their requirements and budget. Google has also addressed problems that can appear when creating extremely wide or tall images, including visual tiling artefacts.

Another important feature is the model’s ability to use information from Google Search and Image Search to improve generated visuals. This grounding capability can help the model work with current or specific visual information rather than relying entirely on its existing knowledge.

Nano Banana 2.1 also supports video-to-image capabilities, opening up additional possibilities for creators who want to turn moments from video into new visual assets.

The technology is being made available through Google’s Gemini ecosystem, including Gemini, Google AI Studio and the Gemini API. Developers can access the model using the model identifier gemini-nano-banana-2.1.

Google is positioning Nano Banana 2.1 as part of a broader family of image-generation models. Nano Banana Pro remains aimed at more advanced visual tasks, while Nano Banana 2.1 focuses on balancing quality, speed and cost. A lighter version is also positioned for users who prioritise speed and lower computing requirements.

The launch comes as competition in AI image generation intensifies. Technology companies are increasingly trying to improve not just how realistic AI images look, but how reliably they can be produced at scale.

That shift is important for businesses. A creative team may want to generate multiple versions of a product advertisement, an online retailer may need different images for thousands of products, while a publisher could require illustrations in several formats. Consistency and cost can matter just as much as image quality in such cases.

Google’s latest model reflects that changing demand. The focus is moving beyond simply asking AI to “create a picture” towards giving users more control over characters, products, text, composition and editing.

The lower pricing may ultimately prove just as significant as the visual improvements. If AI-generated imagery becomes cheap enough to use routinely, it could change how companies approach design and content production.

Nano Banana 2.1 therefore represents another step in Google’s effort to make generative AI imagery more useful outside the novelty phase. Its combination of lower costs, stronger consistency, improved text rendering and broader editing capabilities could make AI-generated visuals easier to integrate into everyday creative and commercial work.

 

Categories
Technology

Google brings Fitbit Air to India

Google has launched the Fitbit Air in India, adding a new screen-free option to the country’s growing wearable technology market. Priced at ₹13,999, the fitness tracker is designed to monitor health, activity and sleep continuously without putting a display or notifications on the user’s wrist.

The Fitbit Air will be available in India from October 2 through the Google Store and selected retail partners. It comes in four colours — Obsidian, Berry, Lavender and Fog.

Unlike a smartwatch or conventional fitness band, the Fitbit Air has no screen. It is built around a small, lightweight tracker that sits inside a wearable band. Users can check their health and fitness information through the companion Google Health app on a smartphone.

The screenless design is central to Fitbit Air’s pitch. Instead of encouraging users to check notifications, messages or statistics throughout the day, the device quietly collects health data in the background. The approach is aimed at people who want continuous fitness tracking without another screen competing for their attention.

The tracker monitors several health and fitness indicators. These include continuous heart rate, resting heart rate and heart rate variability. It also supports blood oxygen saturation, or SpO2, monitoring and irregular heart rhythm notifications associated with atrial fibrillation.

Sleep tracking is another major feature. Fitbit Air can monitor sleep duration and different sleep stages, including light, deep and REM sleep. The lightweight design also makes it suitable for overnight use, allowing users to collect sleep data without wearing a larger smartwatch.

Google is also using the Fitbit Air to expand its personalised health and fitness services. The device works with the company’s Health Coach, which analyses information collected through Fitbit and provides personalised guidance based on activity, sleep, recovery and other wellness data.

The Health Coach is designed to turn raw fitness data into more practical recommendations. Instead of simply showing how many steps a person has taken, the service aims to help users understand broader patterns and make changes to their exercise and recovery routines.

Workout tracking is another part of the experience. Fitbit Air can automatically identify and record certain activities. Google is also introducing features that use a smartphone camera to help users capture information about workouts and exercise equipment.

Battery life is one of the device’s key selling points. Google claims that Fitbit Air can last for up to seven days on a single charge. A quick five-minute charge can provide up to a day of use, making the tracker less dependent on frequent charging than many smartwatches.

The wearable uses interchangeable bands, giving users options for different activities and occasions. The Performance Loop comes with the tracker, while additional band choices include an Active Band designed for workouts and an Elevated Modern Band for everyday wear.

Indian buyers will also receive a three-month trial of Google Health Premium with the device. The premium service provides access to the complete Health Coach experience and additional personalised wellness features. Users will need to subscribe after the trial period to continue accessing those premium features.

At ₹13,999, however, Fitbit Air sits at a relatively high price point for a device that does not have a display. Its pricing could become an important consideration for Indian consumers, particularly those comparing it with conventional fitness bands, smartwatches and other screen-free health trackers.

The product enters a category that has gained attention globally through devices such as Whoop, which have popularised screen-free health and fitness tracking. Fitbit Air follows a similar concept but combines it with Google’s Fitbit ecosystem and Health Coach platform.

The difference between a smartwatch and the new Fitbit is therefore quite clear. A smartwatch puts notifications, apps, calls and health information directly on the wrist. Fitbit Air removes those functions and concentrates primarily on collecting health and fitness data.

The device can also complement a smartwatch rather than necessarily replace one. Users could wear a smartwatch during the day and switch to Fitbit Air for sleep or recovery tracking. Its smaller form factor could make continuous wear more convenient for some users.

Fitbit Air is compatible with both Android and iOS devices, giving it a wider potential audience beyond Google’s own smartphone users.

The launch comes as India’s wearable market continues to evolve. Consumers are increasingly looking for more detailed information about sleep, recovery, heart health, workouts and daily activity. At the same time, some users are becoming more conscious of the number of screens and notifications they interact with every day.

Fitbit Air attempts to address both trends with a single device. It collects detailed health information while deliberately leaving the display behind.

For Google, the launch strengthens Fitbit’s presence in India’s competitive wearable market and expands its portfolio beyond conventional smartwatches and fitness bands. For consumers, the product offers a different approach to wearable technology, focused on continuous health tracking rather than constant interaction.

The success of Fitbit Air in India will ultimately depend on how consumers weigh its screen-free design, health monitoring features, AI-powered coaching and seven-day battery life against its ₹13,999 price and subscription-based premium services.

 

Categories
Technology

Google adds voice, avatars to Gemini

Google is making its Gemini artificial intelligence platform more conversational with new voice and avatar capabilities that allow AI-generated characters to speak, express themselves and interact with users in real time.

The latest rollout includes Gemini 3.8 Flash TTS, Gemini 3.8 Flash-Lite TTS and Gemini 3.8 Live with Live Avatar. Together, the updates expand Gemini beyond text and give developers new ways to build voice-based and visual AI experiences.

The new technology is aimed at making interactions with artificial intelligence sound and feel more natural. Instead of simply converting written text into speech, Google’s latest models can control the style and delivery of a voice, generate conversations between multiple speakers and, in some cases, reproduce a voice from a short recording.

The most notable addition to Google’s text-to-speech technology is voice cloning.

Gemini 3.8 Flash TTS can create a synthetic version of a person’s voice using a short audio sample. Reports on the rollout say the system can clone a voice from about 30 seconds of recorded speech.

Developers can also describe the voice they want through natural-language instructions. They can specify characteristics such as the speaker’s style, tone, accent or role, allowing them to create voices suited to different applications.

The technology could be used for digital assistants, audiobooks, games, video content and virtual characters. A company, for example, could create a branded AI assistant with a consistent voice, while content creators could generate narration without recording every line themselves.

Google has also designed the models to produce more expressive speech. The AI can adjust elements such as tone, pacing and delivery, helping conversations sound less robotic.

Gemini’s new text-to-speech capabilities are not limited to individual sentences.

The models can generate dialogue involving multiple speakers, allowing developers to create AI conversations with different voices. This could make the technology useful for podcasts, educational content, interactive stories and other applications where dialogue is central.

The system is also designed to handle conversational flow more naturally. Rather than producing isolated audio clips, developers can use the technology to create a complete exchange between AI-generated speakers.

Google is offering Gemini 3.8 Flash-Lite TTS as a lighter option for applications that need speech generation at scale. This could benefit developers building voice agents and services that need to produce large amounts of audio.

The new models are being made available through Google’s developer tools, including Google AI Studio and the Gemini API.

Google is also changing how its conversational AI looks.

With Gemini 3.8 Live with Live Avatar, users can interact with an animated AI character that speaks and responds during a live conversation.

The avatar can synchronise its mouth movements with generated speech and use facial expressions while responding. This creates a visual layer on top of Gemini’s existing conversational abilities.

The system is designed for near real-time interaction, allowing users to communicate with an AI character rather than simply receiving a text or audio response.

Google is initially focusing Live Avatar on enterprise applications. Businesses could use the technology for customer support, digital guides, training, product demonstrations and interactive kiosks.

Companies can select existing avatars or create their own digital characters with specific appearances and voices. The technology can be integrated into websites, mobile applications and other interactive experiences.

The ability to create realistic voices and faces also raises concerns around synthetic media and identity.

Voice cloning can make it difficult to distinguish between genuine recordings and AI-generated audio, particularly when a person’s voice is reproduced from a short sample.

Google says its AI systems include safeguards designed to address these risks. The company is also using SynthID, its watermarking technology for identifying AI-generated content, with the Live Avatar system.

The safeguards are particularly relevant as AI-generated voices and faces become increasingly realistic and accessible to developers.

The latest updates show Google’s broader push to make Gemini a multimodal AI platform.

The company’s AI assistant has already expanded beyond text to work with images, audio and other forms of information. The latest developments focus on making the output itself more natural, particularly when Gemini is used for conversation.

The new text-to-speech models give developers greater control over how an AI system sounds. Live Avatar adds another layer by giving that system a face and visible expressions.

For consumers, the technology could eventually lead to more natural digital assistants and interactive AI characters. For businesses, it opens up applications in customer service, training, entertainment and digital communication.

Google’s latest Gemini rollout therefore focuses not only on what artificial intelligence can say, but also on how it speaks, how it sounds and how it appears while communicating.

With voice cloning, expressive speech and animated avatars now part of the Gemini ecosystem, Google is pushing conversational AI closer to a more human-style interaction while continuing to build safeguards around increasingly realistic synthetic media.

 

Categories
Beyond

AI giants face lawsuit over alleged slowdown pact

Four of the biggest names in artificial intelligence, Anthropic, OpenAI, SpaceXAI and Google,  are facing an antitrust lawsuit in the United States over allegations that they agreed to slow the pace of AI development.

The lawsuit, filed in the US District Court for the Northern District of California, was brought by four people who pay for ChatGPT, Claude, Grok or Gemini. They are seeking to represent a wider group of paid subscribers across the country.

The case is a growing debate inside the AI industry: how quickly should increasingly powerful AI systems be developed, and can rival companies work together to make that process safer without breaking competition laws?

The plaintiffs allege that the four companies crossed an antitrust line by coordinating efforts to slow their AI development. They argue that if competing companies collectively agree to reduce the pace at which their products improve, consumers could receive less value from the AI services they pay for.

The lawsuit does not challenge the companies’ right to slow their own development independently because of safety concerns. Instead, the plaintiffs argue that competing firms should not agree among themselves to restrain development.

The immediate trigger for the case was a statement by Anthropic CEO Dario Amodei on September 12. Amodei called for cooperation among leading AI companies to slow advances in AI capabilities and give safety measures more time to catch up.

The proposal quickly received public support from OpenAI CEO Sam Altman, SpaceXAI CEO Elon Musk and Google DeepMind co-founder and chair Demis Hassabis. The lawsuit points to those responses as evidence of coordination among major AI rivals.

The plaintiffs also allege that discussions about slowing AI development started earlier. Their complaint points to a statement from July signed by senior employees from several leading AI laboratories. That statement acknowledged the intense competitive pressure companies face when considering whether to slow development on their own.

According to the lawsuit, this creates a difficult competition problem. If one company slows its AI development alone while its rivals continue moving ahead, it could lose customers and market share. The plaintiffs argue that an agreement among several competitors removes that competitive pressure.

The companies, however, have been discussing the issue primarily from a AI safety perspective. The argument from industry leaders is that the technology is advancing rapidly and that safety testing, evaluation and safeguards need to keep pace.

Amodei has warned about the possibility of increasingly capable AI systems being used in cyberattacks, biological threats and other harmful activities. He has argued that leading AI developers should have more time to evaluate increasingly powerful models before pushing them further.

The Anthropic chief also acknowledged that cooperation between competing companies could raise antitrust concerns. He suggested that the US government could help facilitate discussions or provide a limited legal framework for certain safety-related cooperation.

Altman has separately supported the idea of a federal framework establishing consistent AI safety requirements. OpenAI has said such a framework could provide greater confidence around safety work, while arguing that companies should not necessarily have to wait for new legislation before working on safety measures.

That tension is now at the centre of the lawsuit. US antitrust law is designed to protect competition and prevent businesses from making agreements that improperly restrict the market. The legal question will include whether safety-related cooperation between competing AI companies can be treated differently from an agreement that limits competition.

The case also raises questions about the future of AI regulation. The technology industry is under pressure to address concerns about powerful AI models while continuing to invest heavily in new systems. Companies face a difficult balance between moving quickly, keeping up with rivals and introducing stronger safety measures.

The lawsuit could therefore become important beyond the four companies named in the case. A court decision could help clarify how US antitrust rules apply when competing AI companies cooperate on safety standards, testing or development limits.

The plaintiffs are seeking to represent a nationwide class of paying subscribers. Their argument is that customers could be affected if competition between major AI platforms is reduced and improvements in services become slower than they otherwise would have been.

The companies have not immediately responded publicly to the lawsuit in detail, according to reports. The case is at an early stage, and the allegations have not been established in court.

The dispute comes at a time when ChatGPT, Claude, Grok and Gemini are competing for users, businesses and developers. Their rivalry has helped drive rapid advances in generative AI, but it has also increased concerns over safety, regulation and the risks of developing increasingly capable systems.

The lawsuit now puts that wider debate before a US court: whether AI companies can cooperate to address genuine safety concerns while still maintaining the competition that drives innovation.

 

Categories
Technology

Google Gemini AI breached three firms

Google’s Gemini AI model breached the computer systems of three real companies during a cybersecurity test in May, raising fresh questions about the risks of giving increasingly powerful AI systems access to the internet.

Google confirmed the incidents on September 18 after reports emerged about the tests. The company said Gemini accessed systems that were outside the intended testing environment but stopped its activity after recognising that the targets were real companies.

The incidents happened during a cybersecurity evaluation conducted by Irregular, an AI security testing company. Gemini was taking part in a “capture the flag” exercise designed to test how effectively an AI model could identify and exploit security weaknesses.

The exercise was supposed to take place inside a controlled environment using fictional companies. However, Gemini was unintentionally given internet access. One of the fictional companies also had the same name as a real company, allowing the model to find and interact with the real organisation online.

The model then moved beyond the boundaries of the exercise. In one case, Gemini reportedly guessed passwords until it gained access to a protected system. In two other cases, it found credentials that had been exposed in public online repositories and used them to enter protected systems.

Google said Gemini stopped each intrusion once it realised that it had accessed real companies rather than the simulated targets it had been instructed to work on.

The names of the three companies have not been disclosed. Google said the affected organisations were informed about the incidents. It also worked with Irregular to change the testing process and address the security problems that allowed the model to move outside the intended environment.

Heather Adkins, Google’s vice-president of security engineering, said the incidents showed why powerful AI models need to be trained to behave responsibly when operating with access to digital systems.

The company did not immediately make the incidents public. Google said it initially decided against disclosure because Gemini had caused no reported damage and had stopped the activity on its own after identifying the mistake. The incidents became public after journalists asked Google about them.

The episode is significant because it is the first known case in which Google has disclosed a Gemini model independently accessing and breaching real third-party systems during testing.

It also comes amid a growing list of similar cases involving other major AI developers. OpenAI, Anthropic and Meta have all disclosed cases linked to cybersecurity evaluations in which AI models moved beyond the boundaries of simulated tests.

Irregular said the Google incident was linked to the same testing problem behind some of the earlier cases. The company said relevant AI laboratories were informed in July and that the known problems with its testing process had been fixed.

The incidents highlight a challenge facing the AI industry. Newer AI agents can do much more than generate text. They can browse the internet, write code, search databases and interact with computer systems. When such capabilities are combined with inadequate safeguards, an AI model can potentially take actions that its developers did not intend.

The Gemini incident also shows how easily a testing error can create an unexpected pathway into the real world. A test designed around fictional targets became connected to genuine companies because of internet access and a naming overlap.

Security researchers have increasingly called for stronger isolation during AI cybersecurity testing. Test environments need to prevent models from reaching real systems, while credentials and other sensitive information must be kept away from AI agents during evaluations.

The latest situation does not indicate that Gemini carried out a deliberate attack against the three companies. Google has said the model believed the systems were part of its assigned test and stopped once it recognised the mistake.

This still adds to concerns about AI safety, autonomous AI agents and loss of control. Researchers are closely watching how models behave when they are given greater freedom to plan and execute tasks without constant human intervention.

Google said it has worked with its testing partner to strengthen the process. Irregular has also said that it has resolved the known issues and is working on safer practices for AI cybersecurity evaluations.

The Gemini episode therefore serves as another warning for the AI industry. The more access AI agents receive, the more important it becomes to keep testing environments isolated, credentials protected and clear limits in place.

 

Categories
Technology

Google tests paying publishers for AI answers

Google is testing a new way to pay publishers whose content helps power its artificial intelligence responses, marking a significant shift in how the search giant may value online content in the AI era.

The programme, called the AI Contribution Pilot, is being offered to a limited group of publishers through Google Search Console. Participating websites receive a new earnings section showing how much they have earned from their content contributing to responses across Google AI Overviews, AI Mode and the Gemini app.

The pilot is still in its early stages and Google has not opened it to publishers generally. Industry reports indicate that dozens of publishers have been approached, with the programme appearing to attract particular interest from small and mid-sized publishers. The participating websites are not limited to news organisations, suggesting that Google is testing a broader model for compensating websites whose information is used by its AI systems.

The basic idea is simple: when a publisher’s material makes a meaningful contribution to an AI-generated answer, Google can assign value to that contribution and make a payment. This is different from the traditional search model, where publishers generally depend on users clicking a Google result and visiting their websites to generate advertising or subscription revenue.

Google has described the initiative as an early-stage learning pilot designed to explore how it can reward high-quality content while continuing to provide publishers with traffic and tools. The company has also said it wants to work with websites whose material helps keep its generative AI answers fresh and accurate.

There is, however, an important catch. Publishers can see a monthly earnings figure in Search Console, but Google currently provides little information about how that amount is calculated. The absence of a clear payment formula has already raised questions among publishers about transparency and whether the amounts being offered properly reflect the value of their content.

The system is therefore closer to a pay-per-value or usage-based AI licensing model than a conventional licensing agreement with a fixed annual fee. Publishers do not appear to receive a large upfront payment. Instead, earnings are linked to how Google assesses the contribution of their content to AI responses. Participants can also opt out of the programme.

That distinction matters because Google’s AI products are changing how people consume information online. A user who once searched for a question, opened several websites and read their articles may now receive a complete answer directly on Google through an AI Overview or AI Mode. The change can save users time, but it also creates a difficult business problem for publishers.

Publishers spend money on reporters, editors, researchers, photographers and technology to create the information that AI systems rely on. If users receive the substance of that work without visiting the original website, publishers can lose valuable search traffic, advertising opportunities and potential subscribers.

Recent research has added weight to those concerns. A field experiment published in August found that removing Google’s AI Overviews and AI Mode increased click-through rates to publishers, while an AI Mode-only experience reduced referrals. The findings underline the growing tension between convenient AI search and the economics of the websites that supply the information behind it.

Google has been expanding its measurement tools at the same time. Its Search Console now includes reporting around visibility on generative AI surfaces, giving websites more information about how their content appears in Google’s AI experiences. The new payment pilot takes that relationship a step further by attaching a financial value to some of that AI exposure.

The development also comes as Google faces growing regulatory and legal pressure over the use of publisher content in artificial intelligence. European regulators have been seeking views from publishers on Google’s AI search opt-out system, which allows websites to prevent their content from being used for AI-generated summaries without affecting their traditional search rankings.

The payment experiment could therefore serve several purposes at once. It gives Google a way to test how a future AI content licensing market might work, while giving publishers an opportunity to receive some compensation for material that contributes to AI answers.

Some publishers see that as an important first step. Even if the current payments are small, the fact that Google is testing direct compensation could establish a new precedent: content used by AI systems can have measurable economic value.

Others remain sceptical. Without greater transparency around the calculation of payments, publishers may find it difficult to determine whether the programme offers a fair return. Some industry executives have reportedly described early payouts as too small compared with the advertising revenue their content generates.

 

Categories
Technology

Google Cloud brings Gemini AI to legal work

Google Cloud has launched Gemini Enterprise for Legal, a specialised artificial intelligence platform designed to help law firms and corporate legal departments automate time-consuming legal work while keeping sensitive information within controlled environments.

Announced on August 25, the new offering brings AI agents into workflows such as contract review, legal research, regulatory monitoring, privacy requests and document drafting. Unlike general-purpose AI tools that mainly respond to prompts, Google says its new platform is designed to carry out multi-step tasks using a firm’s own systems, data and working rules.

The service is currently available in preview and has been introduced with the involvement of major law firms including Cleary Gottlieb, Freshfields, Weil and Williams & Connolly. Google is positioning the product as part of a wider push to build industry-specific AI solutions rather than relying on a single general-purpose model for every professional task.

A key difference is the emphasis on the particular demands of legal work. Lawyers often deal with privileged client information, confidential documents and separate access rights for different matters. They also need research to be based on reliable legal authority rather than simply generated from an AI model’s training data.

Gemini Enterprise for Legal is designed to address those concerns by connecting with a firm’s existing document and matter-management systems. Google says the platform can inherit existing permissions and ethical walls, helping prevent information from one client or matter from being accessed inappropriately by another. Client files, internal playbooks, negotiated positions and other organisational data are also intended to remain within the firm’s private data environment.

The platform has four main elements. The first is a collection of purpose-built legal skills that guide AI agents through tasks such as contract review, legal brief drafting, citation verification, regulatory monitoring and Data Subject Access Request, or DSAR, processing.

The second is a network of secure connectors that allows Gemini Enterprise for Legal to work with software already used by legal teams. These include iManage, NetDocuments, DocuSign, Everlaw, RelativityOne, Thomson Reuters, Harvey, Legora and CourtListener, among others. Google says the integrations are designed to preserve existing access controls instead of requiring firms to move their data into a separate system.

The third element is an ecosystem of third-party legal technology providers and AI agents. Companies and consulting firms including Accenture, Deloitte, KPMG and others are working with Google to extend the platform and build customised capabilities.

The fourth is the underlying Gemini Enterprise platform, which provides a central control system for IT and risk teams. It includes governance, audit logging and risk-management features, giving organisations greater visibility into how AI is being used.

The practical appeal for lawyers could lie in the amount of routine work the system is designed to handle. Gemini Enterprise for Legal can monitor legislative changes, court dockets and regulatory developments, then compare those changes with an organisation’s policies and flag potential areas of exposure.

Contract work is another major focus. The AI can examine vendor agreements, non-disclosure agreements and merger-and-acquisition documents against a firm’s existing playbooks. It can highlight clauses that may create risk, allowing lawyers to spend more time on negotiation and professional judgment rather than first-pass document review.

The platform can also turn older contracts into structured contracting playbooks by identifying commonly used terms, fallback positions and other institutional knowledge. That could help firms maintain consistency across large numbers of agreements while reducing the manual effort involved in building and updating internal guidance.

Privacy teams are another target. The system can search across enterprise systems to locate personal information needed for DSAR responses, potentially reducing the time spent collecting information manually and helping organisations meet regulatory deadlines.

Other applications include preparing and redacting documents for court filings, as well as drafting NDAs while following a firm’s preferred structure and formatting. Google describes this shift as a move from AI that simply answers questions to agentic AI that can execute defined tasks from beginning to end, with human professionals remaining responsible for critical decisions and review.

Security is central to Google’s pitch because legal technology involves some of the most sensitive corporate and personal information. Google says client data, firm-specific playbooks, intellectual property, custom agents and model outputs remain private to the organisation and are not used to train or fine-tune its foundation models.

The launch also highlights the intensifying competition in legal AI. Technology companies and specialist legal-tech firms are increasingly targeting law firms as demand grows for faster research, document analysis and workflow automation. Google’s strategy is to work alongside existing legal software rather than force firms to replace their current systems.

For law firms, the bigger change may therefore be less about replacing lawyers and more about changing how legal teams divide their time. Routine searches, document checks and compliance monitoring can increasingly be delegated to AI agents, while lawyers concentrate on interpretation, strategy, negotiation and client advice.

Gemini Enterprise for Legal is still in preview, meaning its capabilities and integrations will continue to evolve. Google has also indicated that more industry-specific versions of Gemini Enterprise are planned, signalling a broader strategy of taking enterprise AI beyond general productivity tools and adapting it to specialised professional workflows.

 

Categories
Technology

Google brings Gemini 3.7 Flash for coding, agents

Google has launched Gemini 3.7 Flash, its latest artificial intelligence model aimed at making software development, coding and AI-agent workflows faster, more accurate and less expensive. The new model, unveiled on August 13, arrives only three weeks after Gemini 3.6 Flash and is positioned as Google’s most capable “workhorse” model so far for coding and agent-based tasks.

The release reflects Google‘s growing focus on AI agents that can do more than simply generate text or answer questions. Gemini 3.7 Flash is designed to handle multi-step tasks, use software tools, work through complex instructions and assist with workflows that previously required substantial human intervention.

Google said the model brings improvements across software engineering, web development and knowledge-intensive work, including areas such as finance, law and biosciences. The company also highlighted better debugging, issue resolution, code generation and instruction-following.

One of the biggest changes is the model’s ability to produce better code on the first attempt. Google said Gemini 3.7 Flash achieved a 43.6% score on the FrontierCode 1.1 Main benchmark, compared with 34.4% for Gemini 3.6 Flash. On DeepSWE v1.1, which measures software-engineering capabilities, the new model scored 65.3%, up from 49% for its predecessor.

For developers, that improvement could mean fewer rounds of prompting, debugging and manual correction. Google said Gemini 3.7 Flash is better at dealing with roadblocks, understanding when clarification is required and following instructions more closely. It also puts more effort into multi-step planning and tool calls, potentially reducing failed attempts when an AI agent is working through a complicated software task.

Web development is another major focus. Google said Gemini 3.7 Flash can create more functional layouts and feature-complete applications with fewer prompts. It can also work from screenshots, images or complete design systems and produce interfaces that more closely follow the original design.

The model recorded an Elo score of 1,588 on Arena.ai’s WebDev Arena, compared with 1,538 for Gemini 3.6 Flash. That improvement is significant for developers using generative AI to create websites and applications, where producing working code is only part of the challenge. Matching the intended user interface and visual design has become equally important.

Google is also pitching Gemini 3.7 Flash as a model for knowledge work and business automation. On the GDP.pdf benchmark for complex document understanding, it scored 34%, compared with 22% for Gemini 3.6 Flash. On AutomationBench, which tests real-world business workflows, the new model achieved 30.4%, compared with 17% previously.

The pricing is another important part of the launch. Google has introduced Gemini 3.7 Flash at $0.75 per million input tokens and $3.75 per million output tokens until the end of 2026. From January 1, 2027, those prices are scheduled to rise to $1.50 per million input tokens and $7.50 per million output tokens.

The lower introductory price is significant because running AI agents can consume large amounts of tokens. Agents often require repeated model calls as they plan tasks, access tools, inspect results and make corrections. Lower inference costs can therefore make autonomous AI systems more practical for businesses and developers operating them at scale.

Gemini 3.7 Flash is available through several Google platforms, including the Gemini API, Google AI Studio, Google Antigravity and Android Studio. Enterprise customers can access it through the Gemini Enterprise Agent Platform and Gemini Enterprise app. The model is also being incorporated into Google’s consumer-facing Gemini Spark service for eligible Google AI Pro and Ultra subscribers.

Gemini Spark, described by Google as a personal AI agent, can use the new model to perform multi-step knowledge-work tasks. Google said the upgrade improves its use of Google Workspace applications, allowing Spark to consolidate files, draft emails and update status documents while working under user direction.

The launch also underlines how quickly the generative AI market is evolving, with technology companies competing across AI models, coding tools and autonomous agents. Google’s three-week gap between Gemini 3.6 Flash and Gemini 3.7 Flash shows the pressure on AI companies to improve models rapidly while keeping inference costs under control.

Google is competing in an increasingly crowded market that includes OpenAI and Anthropic, particularly in coding assistants and autonomous AI workflows. The company is simultaneously developing its premium Gemini models, with its next flagship Pro release still being closely watched by the industry.

Google has also included updated safeguards covering cyber misuse and chemical, biological, radiological and nuclear-related risks. The company said the new model was developed with these protections in mind as it expands the capabilities of AI agents.

With Gemini 3.7 Flash, Google’s message is clear: the next stage of AI competition will not be judged only by how intelligently a model answers a prompt, but by how reliably, affordably and independently it can turn that prompt into completed work.

Categories
Leaders

Google AI veteran Jeff Dean leaves to build startup

Jeff  Dean, Google’s longtime chief scientist and a key architect of its artificial intelligence efforts, has left the company to co-found Discovery Loop, a new AI startup focused on using artificial intelligence to accelerate scientific and engineering breakthroughs.

Dean is not leaving alone. He is joining forces with three other prominent Google AI researchers, Sanjay Ghemawat, Oriol Vinyals and Quoc Le, in a move that brings together some of the industry’s most experienced researchers outside Google’s corporate structure.

Discovery Loop is being established as a public-benefit corporation, with a mission that goes beyond building another consumer AI chatbot. The company wants to use AI to automate parts of the scientific method — allowing systems to generate ideas, design experiments, analyse results and repeatedly test new possibilities.

The ambition is to make scientific discovery faster and more scalable.

Dean’s departure is particularly significant because of his extraordinary influence on Google’s technical foundation. He joined the company in 1999, when Google was still a relatively small organisation, and went on to become one of its most respected engineers and researchers.

Over nearly three decades, he worked on systems that helped Google handle enormous amounts of information and built technologies that became important to the company’s search, computing and AI infrastructure.

He also played a major role in Google’s development of specialised hardware for machine learning, including the Tensor Processing Unit (TPU) programme. TPUs later became a crucial part of Google’s strategy for training and running large AI models.

Dean eventually became Google’s chief scientist, putting him at the centre of the company’s long-term technology strategy.

His move comes at a particularly important moment for Google.

The company is reorganising its AI leadership as competition intensifies across the industry. Google is facing pressure from OpenAI, Anthropic and other AI companies to move quickly in areas ranging from frontier AI models and AI agents to scientific research and coding.

Google DeepMind is also undergoing a leadership reshuffle. Demis Hassabis, who has led DeepMind, is moving away from day-to-day executive responsibilities to focus more heavily on long-term research and become chairman and chief scientist. Koray Kavukcuoglu is taking greater responsibility for AI model development.

Dean’s departure is therefore part of a broader period of change inside Google’s AI organisation.

Yet the creation of Discovery Loop also highlights how the next phase of artificial intelligence may extend beyond the race to build increasingly capable general-purpose models.

The startup wants to focus on what could be called AI for discovery,  systems capable of working through complex scientific problems by repeatedly proposing, testing and refining ideas.

The potential applications are wide. The company is expected to explore areas such as drug discovery, hardware design, engineering and other scientific challenges where progress often depends on running large numbers of experiments.

Traditionally, scientific research can be slow because experiments require time, specialised equipment and human researchers. AI could potentially shorten that cycle by helping researchers identify promising ideas, automate parts of experimental work and analyse huge amounts of data.

Discovery Loop’s founders believe this could create a new model for scientific research, where AI systems operate alongside scientists and engineers rather than simply serving as productivity tools.

The company has also attracted significant backing. Its investors include prominent venture capital firms such as Radical Ventures, Khosla Ventures, Lightspeed, Kleiner Perkins and Doerr Capital. Alphabet, Google’s parent company, is also participating as an investor and cloud partner.

That relationship makes Dean’s exit unusual. While Google is losing a senior AI figure, the company is also backing the new venture and providing cloud infrastructure.

The move reflects the changing relationship between large technology companies and AI startups. Major researchers can now leave established companies with deep technical experience, access to capital and ambitious ideas — while their former employers may still have reasons to support their work.

For Dean personally, the move marks a dramatic change after 27 years at one of the world’s most influential technology companies.

He joined Google during its early growth and remained there through the transformation of search, cloud computing, smartphones, machine learning and generative AI.

Now, instead of helping shape the future from inside Google, he is attempting to build a new organisation around one central question: Can AI dramatically speed up the way humans discover new things?

That question could become increasingly important as artificial intelligence moves from generating content to performing increasingly complex tasks.

Discovery Loop’s success will depend on whether its systems can produce reliable, measurable breakthroughs rather than simply promising faster research.

But with four highly experienced AI researchers at the helm and backing from major technology and venture investors, the startup is already attracting attention.

Categories
1 Minute-Read

EU slaps $1 bn fine on Google again

Google has been fined €1.02 billion (around $1.18 billion) by the European Union for violating the bloc’s Digital Markets Act (DMA), marking one of the biggest penalties imposed under the landmark tech regulation.

EU regulators said the company abused its dominant market position by giving preferential treatment to its own services, undermining fair competition. Google said it disagrees with the decision and plans to challenge the fine.

The ruling adds to the company’s ongoing regulatory challenges in Europe and reinforces the EU’s efforts to curb anti-competitive practices among major technology firms and promote fair digital markets.