Meta Shuts AI Cost Tracker After Exposing Model Expenses 

Meta has ended its partnership with Mercor, a $10 billion AI data startup, after a supply chain attack exposed some of the AI industry’s most closely guarded secrets. The breach, which began with a compromised open source library in late March 2026, revealed not only personal data but also the training methods behind top language models. Hackers used a tampered version of the LiteLLM open-source library, leading to investigations at OpenAI and Anthropic and a class-action lawsuit involving over 40,000 people.  

Last month, in March 2026, hackers targeted a popular open-source library and stole more than just personal data. Wired reports that they may have also taken the blueprints for building some of the world’s most advanced AI models.  

This disruption follows a major cybersecurity breach. Meta has put its partnership with Mercor, a San Francisco AI data company, on hold after a March 5, 2026, cyberattack exposed private details about Mercor and possibly its clients. The pause has no set end date and has made many in the industry nervous since companies have spent billions developing these Secret methods.  

The Startup Behind the Curtain 

Mercor may not be widely known, but it plays a key role in the AI industry. Founded in 2023 by Brendan Foody, Adarsh Hiremath, and Surya Midha, three friends from the Bay Area, the company brings together contractors, engineers, lawyers, doctors, bankers, and journalists to create high-quality training data for AI labs. Its clients include Meta, OpenAI, Anthropic, and Google.  

Mercor’s growth has been remarkable, even in October 2025, for Silicon Valley. It raised $350,000,000,000 in a Series C round, reaching a $10,000,000,000 valuation and making its three founders the world’s youngest self-made billionaires at 22. By September 2025, the course’s annual revenue hit $150,000,000 up from $100,000,000 just three months ago. Its focus on creating fine-tuning and reinforcement learning data for AI labs has made it one of the most valuable private companies in the AI supply chain.  

However, Mercor’s position at the center of the air supply chain has come with risks.  

A Poison Package, a Cascade of Exposure 

The attack on Mercor started further up the supply chain in late March 2026. Wiz, Synk, and Datadog Security Labs found that a hacker group called TeamPCP broke into the CI/CD pipeline of LiteLLM, an open-source Python library. LiteLLM is used by millions of developers and has 97,000,000 monthly downloads, and is found in about 36% of cloud environments.  

Earlier, TeamPCP used a supply-chain attack against Trivy, a popular security scanner, to steal traditional credentials. On March 27, 2026, a light ML LLM maintainer used these credentials to upload two malicious versions of LiteLLM 1.82.7 and 1.82.8 to the Python Package Index (PyPI). The harmful packages were available for about 40 minutes before they were found and taken down.  

The attack was complex. Version 1.8.7 hit base64-encoded malware in the library’s proxy server code, which ran as soon as it was imported. Version 1.8.2 used a harmful path configuration file that triggered every time a Python process started. Both versions were made to steal environment variables, API keys, SSH keys, cloud credentials for AWS, Google Cloud, and Azure, CI/CD secrets, and database authentication, and to send all the stolen data to a server at app. models.litellm[.]cloud.  

Mercor confirmed it was hit by the attack. Later, they discovered nearly 4 terabytes of exposed data. This included platform source code, a large user database, and video interviews with identity documents. The breach may have revealed the names and social security numbers of over 40,000 contractors and customers.  

The Secrets That Matter Most 

The exposure of personal data is serious, but Meta and other AI labs are more concerned about leaked proprietary information.  

Because Mercor manages data pipelines for multiple AI companies, the bridge may have developed expertise and confidential data-processing and training strategies over the years of investment. Unlike datasets, these methods have critical advantages. Several AI labs are investigating the extent of the leak.  

OpenAI is reviewing the incident but hasn’t stopped working with Mercor. Anthropic hasn’t commented publicly. Google is also believed to be assessing the impact of the breach.  

This bridge reveals a major industry risk column when many companies rely on a single supplier. A single bridge can compromise the top AI methods across the sector.  

Extortion and Legal Fallout 

The hacker group Lapsus$, known for past attacks on big companies, claimed responsibility for the NORCO breach and started selling the stolen data on dark web forums. Security experts believe Lapsus$ is working with the team PCP, which has become a major threat in the AI and enterprise software worlds. This group is also believed to be behind a series of supply chain attacks that hit over 1,000 enterprise SaaS environments, including a breach of the European Commission linked to the same campaign by CERT-EU.  

On April 1, 2026, plaintiff Lisa Gil, a resident of Wahiawa, Hawaii, filed a class action complaint against Mercor.io Corp. in the U.S. District Court for the Northern District of California. The suit alleges that Mercor failed to maintain adequate cybersecurity protections, leaving more than 40,000 people exposed to identity theft and fraud. The complaint states that a LiteLLM incident on March 27 was the entry point. It also claims that Mercor’s reliance on a compromised open-source dependency, without sufficient monitoring, created dangerous conditions that led to the breach.  

Meta has not made any public statements about the breach. In March 2026, the company signed a $27,000,000,000 AI infrastructure deal with Nebius Group and experts, spending between $135,000,000,000 and $150,000,000,000 this year, making its air training pipeline extremely important. Stopping work with a T beta vendor is a decision Meta would make only if their secret methods outweighed the cost of hunting operations.  

A Cautionary Tail For The AI Supply Chain 

The Mercor highlights how modern supply chain attacks can expose both credentials and unique intellectual property when AI companies depend on the same data vendors and open-source tools.  

Security companies have warned about this exact problem. Aikido Security, which became a unicorn in January 2026, is based on the idea that open-sourced dependency risk is a major threat to enterprise software. The Mercor breach shows that this risk may be even greater for the AI training pipeline.  

The next few months of 2026 will show whether Mercosur’s rapid growth can continue after a March breach that compromised both user data and clients’ most closely guarded secrets. The AI industry’s rapid pace in 2025 was driven by the belief that this infrastructure was secure. Now, that belief is being questioned.  

Source:  Meta freezes AI data work after breach puts training secrets at risk  

Caltech Enables Micro-metal Printing for Next Gen Wearables 

Caltech researchers have developed a novel 3D printing technology called Hydrogel Infusion Additive Manufacturing (HIAM). This process enables the fabrication of microscale, intricately detailed metallic structures much smaller than previously possible. It could advance the development of high-performance wearable technologies, biomedical devices, and microelectromechanical systems (MEMS).  

Highlights Of Caltex Micro-Metal Printing 

  • HIAM begins by additively manufacturing a hydrogel template, which is subsequently infused with metal precursor ions such as copper or nickel.  
  • The construct then undergoes dual-stage thermal processing. This eliminates the polymer matrix, sinters the metal, and isotopically shrinks the component by up to 90%, yielding dense, miniaturized, and highly precise features.  
  • Microscale metallic components produced in this manner often exhibit nanoscale porosity and grain boundary networks. Surprisingly, these microscopic structural defects yield parts that are three to five times stronger and more fatigue resistant than those of conventional alloys.  

How This Affects Wearables and Smart Technology 

  • This process facilitates the fabrication of miniaturized metallic devices, such as microelectromechanical sensors and actuators, which are critical components of microphones in advanced wearable systems.  
  • Caltech is collaborating with Meta Inc to apply this research toward the development of new augmented reality and virtual reality wearable devices.  
  • This technique also enables the fabrication of flexible, reconfigurable electronic devices, including biosensors and physiological monitoring.  

The research group led by experts such as Julia R. Greer has demonstrated print resolution to 150 nanometers comparable to the size of the influenza virion. This capability suggests applications in ultra-precise biomedical implants and sensing platforms.  

These advancements are enabled by a water-based process developed by Caltech engineers, as highlighted in a recent Nature paper published on October 20.  

This new method works with many metals. The additive manufacturing approach is compatible with multiple metals and alloys and permits compositional gradients within a single manufactured part with marginal modification. It presents potential for miniaturized components in microelectronics, transportation, aerospace, thermal management, and medical fields. Manufacturing creates items layer by layer. This allows for shapes that conventional techniques, such as forging, molding, etching, or milling, cannot make. Most current 3D metal printers use lasers to melt metal powders, achieving detail down to about 100 microns, roughly the thickness of two sheets of paper. Resolution refers to the smallest detail the process can produce.  

One challenge is that metals like copper conduct heat very well. Even with a focused laser, the heat spreads and melts powder outside the intended area, reducing the level of detail achievable.  

To overcome these barriers, Caltech’s team, including Max Saccone, Rebecca Gallivan, Daryl Yi, and Kai Narita, took an innovative path. Rather than print materials directly, they used hydrogels as scaffolds for metal-containing liquids. Kai Narita has since founded 3D Architect to commercialize this technology licensed from Caltech.  

“We need to develop a different approach as we cannot rely on heat to construct our structures,” said Saccone.  

Hydrogels are made from flexible polymers. Hydrogels composed of hydrophilic polymer networks remain intact in aqueous environments and are used in products such as soft contact lenses. Ultraviolet photopolymerization induces crosslinking in liquid polymers, forming solid structures in prescribed patterns. Immersing hydrogels in metal ion solutions results in uniform volumetric ion loading. Calcination at 700 to 1,100 degrees Celsius, depending on the metal, combusts the polymer matrix. Since the metal’s melting point exceeds the calcination temperature, metallic integrity is preserved. The hydrogel also causes the entire piece to shrink, resulting in an even smaller metal part. This method allows the team to 3D print pure metals, alloys, and mixed metal systems with features as small as 40 microns, which is less than half the width of a human hair. They have printed structures from copper, nickel, silver, and various metal alloys.  

The hydrogen infusion additive manufacturing process, or HIAM as we coined it, establishes a means to create metallic materials in an entirely new, much more environmentally friendly way at unrivaled precision levels, says Greer, Ruben F., Donna Mettler Professor of Materials, Science, Mechanics, and Medical Engineering, and Director of the Kavli Neuroscience Institute.  

The research received funding from the US Department of Energy, the Resnick Sustainability, the Massa On Foundation, and Caltech’s AI4science initiative. 

Source: New Process Allows 3-D Printing of Microscale Metallic Parts  

NVIDIA CloudXR Streams RTX Graphics To Spatial Devices

NVIDIA CloudXR 6.0 lets you stream high-quality, RTX-powered graphics to devices such as Apple Vision Pro, Meta Quest 3, Pico for Ultra, and web browsers. As a universal OpenXR bridge, it delivers real-time, photorealistic rendering from remote workstations or cloud servers, giving you untethered, high-quality XR experiences.  

Key Features and Benefits of CloudXR 

  • CloudXR 6.0 streams high-fidelity content to spatial devices and web browsers, including Apple Vision Pro, Meta Quest 3, Pico, and others.  
  • CloudXR 6.0 provides native support for Vision OS, leveraging dynamic foveated streaming to deliver 4K output with minimal latency while preserving user data privacy.  
  • As a universal bridge, CloudXR 6.0 enables developers to build XR apps once and deploy them across platforms such as iOS, iPadOS, and visionOS.  
  • CloudXR 6.0 shifts computationally intensive rendering from XR devices to high-performance workstations. This enables efficient streaming of complex 3D data assets and reduces on-device processing requirements for headsets.  
  • CloudXR.js enables developers to deliver interactive, GPU-rendered 3D content directly to XR device browsers, such as on Apple Vision Pro, Meta Quest 3, and Pico 4 Ultra, leveraging real-time streaming protocols.  

CloudXR 6.0 lets professionals in disciplines like automotive design and healthcare experience interact with, and work together on complex 3D models from anywhere.  

NVIDIA CloudXR 6.0 is a GPU-accelerated streaming platform that brings high-quality spatial experiences. From powerful GPUs to a wide range of AR, VR, and spatial computing devices, the following section explains how its architecture delivers these experiences. With full OpenXR compliance, developers can build once and deploy to any supported headset or operating system. By removing the need for powerful local hardware, the SDK lets you stream photorealistic digital twins and elaborate simulations to lightweight devices, including native support for Apple Vision OS and easy web access through CloudXR.js.  

How CloudXR Works 

NVIDIA CloudXR 6.0 serves as a universal open XR bridge, offloading processing from the XR device. It delivers photorealistic spatial experiences through three core components:  

  • CloudXR runtime (server): This component runs on Windows or Linux workstations and handles GPU-accelerated rendering and low-latency XR content encoding. It connects RTX-powered applications to the network interface for delivery to client devices.  
  • CloudXR frameworks (client) for Apple platforms. There are two ways to build apps that receive CloudXR streams on visionOS. You can use the Foveated Streaming Framework to stream high-quality OpenXR applications. This system sends top-quality content only where it’s needed, based on the user’s location, to maintain high performance. On iOS or iPadOS, you can use StreamingSession.xcframework to stream OpenXR experiences to iPhones or iPads. Both libraries have similar APIs, making it easy to build cross-platform streaming apps.  
  • CloudXR.js (Web Clients) This JavaScript framework makes browser-based XR easy. Devices such as Apple Vision Pro, Meta Quest 3, Pico 4, Ultra, and other supported platforms can access cutting-edge robotics, Omniverse, and open-edge XR content via WebRTC, with no need to install anything from an app store.  

Get Started With Cloudxr 

CloudXR Runtime Server SDK for OpenXR 

Deploy the essential server-side engine to render, encode, and stream your NVIDIA RTX–powered applications. CloudXR Runtime 6.0 works as a standard OpenXR bridge, so any compliant application from NVIDIA, ISAC, Lab, to custom engines can stream photorealistic content to lightweight wireless clients with ultra-low latency. 

Source: NVIDIA CloudXR  

Meta Launches AI Shopping Using Real-World Vision 

Takeaways 

  • Muse Spark was designed specifically for Meta’s products to enhance user experience, making Meta AI smarter and faster. Over time, it will introduce helpful features that leverage recommendations and content from Instagram, Facebook, and Threads, delivering more tailored results to users.  
  • Our models are growing as expected. News Spark is an early step in our progress, and we are already working on larger models.  
  • Muse Spark is our strongest model so far, directly powering the Meta AI app and website for faster, smarter systems. Soon, it will expand into WhatsApp, Instagram, Facebook, Messenger, and AI Glasses, helping users wherever they are. Select partners will also benefit from an improved private API preview.  

Today, we are introducing Muse Spark, the first in a new series of large language models from Meta Superintelligence Labs. Muse Spark aims to become a personal aid, making daily life easier for anyone, anywhere, with support focused on what matters most to users.  

A New Model: Muse Spark  

In the past nine months, Meta Superintelligence Labs rebuilt our AI stack from scratch, moving faster than ever before. Muse Spark is the first in our new Muse series, where each version builds off the last in a careful, scientific way. The first model is small and fast by design, but it can still handle complex questions in science, math, and health. This provides a strong starting point, and we are already working on the next version. News Spark now powers the Meta AI assistant in the Meta AI app and on meta.ai, supporting intricate reasoning and multimodal tasks.  

What’s Changed With Meta AI 

You can switch between modes based on your task, and Meta can use several sub-agents at once to help you. (Sub agents are specialized mini programs that focus on specific tasks. ) For example, if you are planning a family trip to Florida, one agent can make the itinerary. Another can compare Orlando and the Keys, and a third can find kid-friendly activities  all at the same time, so you get better answers faster.  

Ask Meta AI: It Understands  

One major improvement is that Muse Spark has strong multimodal perception. It lets Meta AI see and understand what you see not just what you type. Snap a photo of an airport snack shelf, and Meta will list all snacks with the most protein. Scan a product to compare it to others. When you pair it with our AI glasses, Meta AI will better understand your surroundings.  

This multimodal perception is especially helpful for your health. With Muse Spark, Meta AI can answer health questions that involve images and charts. We worked with doctors to help provide useful information on common concerns.  

Muse Spark is great for visual coding, so you can create custom websites and mini-games. Muse Spark’s advanced visual coding lets users create custom websites and mini-games with a single prompt. Users can ask Meta AI to build dashboards for unique events, design retro arcade games, or launch simulators, then easily share them with friends for creative collaboration or entertainment. Fit ideas, style a room, or pick out a gift for someone. Shopping mode draws on inspiration and brand stories from our apps, showing you ideas from creators and communities you already follow.  

When you look up a place or trending topic, Meta AI brings you useful context right in your conversation. You can tap a location to see public posts from locals, ask people, or ask what people are talking about, and get a full view from content and community posts. It’s information from your community right when you need it.  

Gazing Forward 

We are also giving select partners early access to the technology through a private API. (An API, or application programming interface, lets developers connect their own software to Muse Spark so they can test how it works. We hope to open-source future versions of the model.  

This is just the beginning. As we add more features, users will see more visually detailed results in their answers, including reels, photos, and posts with creators credited alongside improved results. We’re introducing stronger safeguards for safety and privacy, enhancing trust for everyone who uses Meta AI in their lives. We are working toward personal superintelligence, an AI that not only answers your questions but truly understands your world because it’s built around it.  

Source: Introducing Muse Spark: MSL’s First Model, Purpose-Built to Prioritize People 

Microsoft Shifts Copilot to Multi-Model Orchestration System

Microsoft has launched its new Copilot Co-Work feature for participants in the Frontier Program. It expands Microsoft 365 Copilot’s role beyond just creating content to managing and executing tasks across business applications.  

Jared Spataro, Microsoft’s chief operating officer for AI at work, shared the update in a blog post. Wave 3 enhancements focused on multimodal AI and longer workflows. Microsoft has integrated Anthropic’s Claude Co‑work, enabling multi‑step tasks in business taps.  

Copilot, co-work, lets users set a goal. The system then creates a plan and organizes tasks across several tools and files. It moves the work forward while people oversee the process. Spataro said this is a move toward AI that can handle a series of actions, not just single requests.  

This feature uses Microsoft’s multi-model approach. It combines Microsoft’s own AI with partners’ models, such as those from OpenAI and Anthropic. Allowing different models to contribute at different workflow steps.  

Microsoft has updated its research tool with multi-model intelligence. A researcher helps users pull together information from different sources. It produces detailed analysis with well‑researched, cited answers. The new critique feature separates content creation from review. It uses models from frontier labs such as Anthropic and OpenAI. One model writes the first response, and another checks and improves it before it is shared.  

Spataro noted that this method improves researchers’ scores on Microsoft’s Draco benchmark by 13.8% across accuracy, completeness, and objectivity.  

There is also a new model council feature. It lets users compare results from several AI models side by side. This gives more insight into how the AI works. It also helps people make better decisions, Microsoft CEO Satya Nadella wrote on LinkedIn.  

You can run multiple models on the same prompt at the same time, so you can see where they correspond and diverge and understand what each other adds.  

This added transparency may help reduce blind spots and make people more confident in using AI for their work.  

Co-pilot Coworks‘ ability to handle multi-step workflows could help customer experience and operations teams manage processes that span different systems, departments, or data sources. It can help coordinate steps to solve complex customer problems, automate routine tasks such as follow-ups, reporting, and scheduling, and maintain consistency by adhering to company data and governance rules. Model evaluation could also help teams review information from multiple AI perspectives. This could enhance trend analysis and support quality assurance by providing greater visibility into the accuracy and consistency of AI-generated outputs.  

At the same time, using enterprise data and governance controls shows the ongoing need for data security and compliance. This is especially true in regulated industries, said Burton Warner, SVP of Enterprise Technology at Capital Group:  

Because Cowork operates on our enterprise data and within our security and risk boundaries. We can experiment, learn, and scale with confidence, which allows us to move faster and focus AI in places where it actually delivers value.  

Shifting AI From Experiments to Embedded Enterprise Workflows 

The frontier rollout of Co-Pilot Co-Work reflects a wider trend in enterprise AI toward systems that combine planning, execution, and appropriate integration across multiple models, integrating external technologies such as cloud, co‑work, and Microsoft. Organizations are moving toward more modular, interoperable architectures that go beyond assistive features toward more embedded operational functions.  

This approach is increasingly being proposed as a way to make AI more reliable and let organizations choose or compare model results based on their needs. The development of multimodal, workflow-oriented systems is likely to influence how enterprise software platforms are evaluated and adopted for deploying AI in business processes. 

Source: Microsoft Copilot Cowork Signals Shift to Multi-Step AI Workflows for Enterprise Users

OpenAI Expands Codex Limits as Agentic Coding Demand Surges

Today, we are introducing a research preview of Codex, a cloud-based software engineering agent that automates the creation of new features, provides direct answers about your codebase, identifies and fixes bugs, and generates pull requests for review. Each function operates in its isolated cloud sandbox, pre-configured with your repository.  

Codex runs on Codex 1, a version of OpenAI O3 designed for software engineering. It was trained with reinforcement learning on real programming tasks in different environments. This helps it write code that fits human style and pull request preferences, follows instructions closely, and runs tests until they pass, ensuring early access and integration for users. We are making Codex available to ChatGPT Pro, Enterprise, and Business users today. Support for Plus and EDU users is coming soon.  

How Codex Works 

You can access Codex from the Chat GPT sidebar to start a coding task. Describe your goal and click Code. For questions about your code, use Ask. Each request launches in an isolated environment with your repository reloaded. Codex analyzes, edits files, and executes commands like test harnesses, linters, and type checkers. Task times range from one to thirty minutes, depending on complexity. Progress is visible in real time.  

When Codex completes a task, it commits changes in its environment and displays terminal logs and test outputs as proof of every step. You can review results, request additional changes, open a GitHub pull request, or add the changes locally. You can also configure code. Products to closely mirror your real development environment.  

Guide codecs with agents.md files in your repository. These files, like README.md, tell codecs how to work in your code, which testing commands to use, and how to follow your project’s practices. Codecs work best when developed by human developers in a well-defined environment, with reliable testing tools and clear documentation.  

In coding tests and internal benchmarks, Codex 1 performs well even without agents.MD files or extra setup.  

Building Safe and Trustworthy Agents 

We’re releasing Codex as a research preview as part of our step-by-step deployment plan. We focused on security and transparency so users can check Codex’s output. This is especially important as Codex models take on more complex coding tasks. Users can review Codex’s work using citations, terminal logs, and test results if Codex is unsure or runs into test failures. It clearly lets you know so you can decide what to do next. It’s still important for users to review and check all code made by the agent before using it.  

Aligning With Human Preferences 

One of our main goals in training Codex 1 was to make HC outputs more closely match human coding preferences and standards than OpenAI. O3 Codex 1 creates cleaner patches that are ready for review and can be added to your workflow right away.  

Preventing Abuse 

It’s more important than ever to prevent harmful uses of AI in software engineering, like creating malware. At the same time, we need to ensure that security measures don’t block legitimate, useful work even when it uses similar techniques, such as low-level kernel engineering.  

To keep codecs both secure and useful, we’ve trained it to spot and block harmful software requests while enabling productive, valuable work. Our updated policies and strict safety checks draw a clear line between the two. Details are in the O3 codex system card addendum.  

Secure Execution 

The Codex agent operates in a secure, isolated cloud environment. It works without internet access and uses only code from GitHub and your preset dependencies. It cannot reach external websites, APIs, or services.  

Early Use Cases 

OpenAI engineers rely on codecs daily to tackle repetitive tasks such as refactoring, renaming, and testing while staying focused on higher-value work. Codecs accelerate feature development, connect systems, resolve bugs, and draft documentation. Teams push productively further by sorting on-call issues, planning daily, and offloading background work. Codecs minimize context shifts and highlight neglected tasks, enabling engineers to move faster and focus on priorities.  

  • Cisco is testing Codex to accelerate project delivery. As early design partners, they shape Codex’s future by using it in real scenarios and providing direct feedback to OpenAI.  
  • Temporal uses codecs to speed up feature development, fix bugs, test, and refactor big code bases while also handling complex tasks. Disks in the background to keep engineers focused and moving faster.  
  • Superhuman deploys codecs for small, repetitive jobs like boosting test coverage and tackling integration issues, so product managers ship updates faster by making code changes, with codecs engineers stepping in only for code review.  
  • Kodiak leverages codecs to build debugging tools, raise test coverage, and optimize code. Advancing its Kodiak driver for autonomous driving, Codex serves as a crucial reference, providing context and version history to enhance the engineers’ understanding.  

Based on current insights, assign clear, well-defined tasks to multiple agents for maximum efficiency. Experiment with diverse tasks and prompts to fully explore the model’s capabilities.  

Updates To Codex CLI 

We recently launched Codex CLI, an open-source coding agent for your terminal. It brings models like O3 and O4 Mini directly into your workflow, speeding up task completion.  

We are now launching a slimmer Codex 1 based on the O4 Mini, custom-built for Codex CLI. This model accelerates CLI workflows, delivering fast Q&A, quick edits, and strong adherence to instructions and style. It’s the CLI default and available in the API as Codex Mini, latest with ongoing updates.  

Connecting your developer account to Codex CLI is now seamless. Simply sign in with your ChatGPT account and choose your API organization. We handle the API key setup, and Plus and Pro users signing in unlock $5 and $50 in free API credits over the next 30 days.  

Codex Availability Pricing And Limitations 

Starting today, Codex is available globally to ChatGPT Pro, Enterprise, and Business User Plus, and EDU access will follow soon. For the next few weeks, Codex is free to try; afterward, we’ll introduce rate limits and flexible pricing buy extra usage as you need.  

Building with Codecs mini: latest. Access it via the Responses API. Pricing is $1.50 per 1K input tokens and $6 per 1K output tokens, with 75% off for prompt caching.  

Codex is still in the early stages right now as a research preview. It doesn’t support features like image inputs for front-end work or mid-task adjustments to the agent’s work. Also, sending tasks to a remote agent takes longer than editing directly. So it may take some time to get used to it. In the future, working with Codex agents will feel more like collaborating with colleagues asynchronously. As the model improves, we expect agents to handle even more complex tasks for longer periods.  

What’s Next?  

G Envision developers focused on what matters with agents handling the rest to maximize productivity to achieve this. We are building codec tools for seamless real-time collaboration and asynchronous task delegation.  

AI tools like Codex CLI now set the industry pace, helping developers code faster. The new asynchronous multi-agent Codex workflow for ChatGPT is now on track to become the industry standard for high-quality engineering.  

Ultimately, real-time pairing and task delegation will converge. Developers and AI agents will seamlessly collaborate on ideas, tools, and daily tasks asking questions, receiving suggestions, and delegating complex work in a unified flow.  

Looking forward, we plan to add more interactive and flexible ways to work with agents. Soon, developers will be able to provide feedback during tasks, collaborate on strategies, and receive updates as work progresses. We also want to connect Codex with more of your favorite tools. Right now, Codex works with GitHub, and soon you will be able to assign tasks from Codex, the CLI, ChatGPT, the Desktop, and even your issue tracker in our CI system.  

AI is driving rapid productivity gains across software engineering, unlocking new possibilities for individuals and small teams. While we are eager for what’s ahead, we collaborate with partners to deepen our understanding of agencies’ impact on workflows and skill development at every level.  

This is just the start, and we are excited to see what you create with Codex. 

Source: ntroducing Codex

SEC Filing Signals Surge in AI-Driven Growth Across Gaming Industry

The latest discovery in SEC filings reveals a clear and significant trend: artificial intelligence is the primary engine of innovation and revenue across the gaming sector. Increasingly, gaming companies are outlining how they are integrating AI into their business strategies, indicating that AI is not just an experiment; it will be fundamental to their future growth. AI is redefining how games are created, sold, and monetised, with everything from dynamic gameplay to personalised experiences to fully automated content creation.  

AI Becomes Central to Gaming Strategy  

According to recent SEC filings, gaming companies are increasingly focusing on implementing AI to improve game development efficiency and player interactions. Developers can use AI to automate production pipelines, generate assets, optimise in-game mechanics, and ultimately deliver a higher-quality gaming experience with fewer resources. In an industry where development costs continue to escalate, improving efficiency through AI will provide a significant competitive advantage for companies that implement it.  

In addition to improving production, AI is changing how games are played. Game developers are implementing machine learning to create games that adapt to players’ behaviour as they play, offering unique challenges and storylines tailored to how they interact with the game. Not only do these technologies improve gameplay experiences, but they also help increase player retention, a key driver of long-term revenue growth in the gaming industry.  

Unlocking New Revenue Models  

The application of artificial intelligence in the gaming industry has become increasingly prevalent to develop new monetisation opportunities that do not rely on either traditional game purchases or in-game purchases. Some examples of AI-based initiatives include dynamic pricing, personalised content, and AI-generated extensions that can generate additional streams of income by continuously evolving your games into platforms.  

AI enables game developers to create in-game items, quests, and narratives tailored to each player’s preferences, thereby improving engagement and monetisation opportunities. Developers and publishers can also apply AI’s personalisation to new revenue opportunities in the billions of dollars for leading game developers and publishers.  

Transforming Game Development  

AI tools have become an essential component of game development today. With an array of automation tools, AI can assist with coding, animation, and testing. For example, using automated systems, developers can create interactive environments, characters, and dialogue, thereby considerably reducing both the cost and time required to produce video games. The benefit here is that game studios can invest more into the creative aspects of their games rather than spending too much energy on repetitive technical tasks.  

AI testing tools also improve quality assurance by automating bug identification, balancing gameplay visually, and simulating how players may interact with the game. Not only does this produce a higher-quality final product, but it also reduces the risk of costly post-launch fixes. As a result, AI testing tools can improve a game’s profitability and player satisfaction.  

Enhancing Player Engagement  

An impactful way AI can improve gamers’ experiences is by offering more engagement by analysing how they behave, their preferences, and how they perform in the game. With this information, AI can deliver customised experiences that help retain users for longer periods.  

Dynamic difficulty adjustment, personalised recommendation systems, and real-time feedback enable players to engage more deeply in the world of gaming. Personalising your gaming experience will allow for your games to maintain high retention rates, which are an important metric for the success of today’s gaming platforms, which rely on continued use.  

Competitive Dynamics in the Industry  

As the gaming industry incorporates AI into its products, competition is accelerating as companies strive to develop and deploy more advanced technology. Companies that implement AI technology effectively will gain a competitive advantage by improving efficiency, enhancing the gaming experience, and developing more successful monetisation strategies through new video game releases.  

SEC filings show that businesses are now approaching AI not only as a tool but also as a way to differentiate themselves from competitors who create video games using traditional methods in a marketplace filled with similar-sounding offerings. As AI becomes increasingly popular among consumers and developers, the disparity between these groups could continue to grow as the industry evolves.  

Data as a Strategic Asset  

Data collection and analysis are essential to success in AI-driven games; they are among the most widely used methods for generating high-quality content in the industry. Companies are using analytics platforms that can process vast amounts of player data, enabling them to make more accurate predictions about player behaviour and create a more personalised gaming experience.  

Due to the high volume of player data being collected, companies can improve their game design and monetisation models and respond to player feedback much faster. These companies need to be very careful to protect their users’ privacy and security; otherwise, they will lose their users’ trust and fail to comply with the laws/regulations governing the industry.  

Challenges and Ethical Considerations  

While AI offers many possibilities in gaming, it also poses significant challenges. The most important issue is the balance between automated and creative input; placing too much emphasis on AI-generated content could lead to a lack of diversity in gameplay experiences. As a result, it is essential that game developers draw on human creativity as their primary source of inspiration when designing behaviour.  

The ethical implications of AI systems also need to be considered, specifically regarding data use, algorithmic bias, and player exploitation, so developers can design AI systems that encourage fair competition among players.  

Investor Perspective and Market Outlook  

More investors are focusing on the impact of AI on the gaming industry’s growth. With AI capabilities, companies can attract investors looking to invest in high-growth segments of the gaming market. Companies that can successfully implement AI may see this investor group come together to invest in them, given their strong capabilities.  

Analysts expect AI to drive future valuations for gaming companies at both the near- and long-term levels. Therefore, it is anticipated that gaming companies will continue expanding their AI programs to compete with other companies in their markets and to attract investors seeking valuable companies.  

Future of AI in Gaming  

In the coming years, AI will play a bigger role in shaping how video games develop. The development of generative AI, real-time rendering technologies, and cloud-based platforms may lead to new types of interactive experiences.  

Some of these interactive experiences could include fully generated worlds generated by AI and adaptive storyline elements that change based on the player’s actions. There will be many innovative opportunities to play and interact with video games that will redefine how people think about video games and what constitutes entertainment, simulation, or social interaction.  

Conclusion: A New Era of Intelligent Gaming  

Recent data from SEC filings provide clear evidence that Artificial Intelligence is driving growth and innovation within the gaming industry. The industry is on the verge of tremendous change as gaming companies continue to invest in AI-led technologies. Companies that can successfully apply AI technologies to their business models will drive the next growth spurt in the industry and create new opportunities for themselves by redefining how games are created.

Source: https://www.sec.gov/ 

Amazon Targets $5B Drone Delivery Network Across US Suburbs

Amazon will move ahead with developing a large-scale drone delivery network for suburban marketplaces in the United States, with expected investments of $5 billion. The establishment of America’s drone network will continue to be based on ongoing research into logistics innovation. The Amazon delivery method will allow for ultra-fast deliveries using autonomous aerial systems. This initiative demonstrates Amazon’s strategy of redefining last-mile deliveries through automation, AI, and advanced logistics infrastructure.  

Reimagining Last-Mile Delivery  

As opposed to using ground transportation and being forced to take longer routes as a result of any or all of those factors above, drones will provide a direct point-to-point delivery system between two points, thus reducing the amount of time it takes for items to reach their final destination, especially in suburban areas where there are moderate distances and where there is less air congestion.  

The vision for this initiative is for Amazon to create a network of drone delivery hubs that can quickly deliver packages to customers after an order is placed. By doing this, Amazon will set a new standard for customers to expect near-instant delivery rather than treating the package as a premium service.  

Project-Based Logistics Innovation  

Amazon’s expansion has been largely driven by its research into automated logistics systems, which have created a network of drones, fulfilment centres, and AI. The purpose of these initiatives is to optimise efficiency.  

In addition, combining AI with logistics technology will enable drones to find the best delivery route, avoid obstacles, and adapt to changing conditions, including weather and airspace restrictions. By leveraging real-time data and predictive analytics, Amazon aims to develop a responsive delivery system that can be deployed at scale across multiple geographic areas.  

Infrastructure and Investment Strategy  

Establishing a nationwide drone delivery network will require a substantial capital commitment to develop an extensive drone infrastructure. Examples of such investments could include manufacturing drones, establishing charging stations for them, implementing control systems, and creating regulatory compliance frameworks. Amazon’s anticipated $5 billion investment reflects the scope of its logistical transformation and its commitment to the long haul.  

Drones will require an infrastructure consisting of local fulfilment centres, launch pads, and sophisticated monitoring systems to ensure they operate safely and efficiently. This type of infrastructure allows for quicker delivery times whilst reducing dependence on large, centralised hubs. This type of distributed infrastructure supports Amazon’s goal of placing inventory closer to its customers.  

Speed and Customer Experience  

The primary advantage of drone delivery systems lies in their ability to deliver packages within much shorter timeframes. Amazon aims to establish 15-minute delivery, marking a major advancement beyond its existing same-day and next-day delivery options.  

The ability to deliver products more quickly results in greater customer satisfaction and creates new opportunities for e-commerce businesses to offer immediate delivery of essential items, groceries, and small consumer products. The ability to receive products within a few minutes of ordering will transform shopping patterns by driving consumers to buy lower-priced items more frequently, thereby increasing use of the shopping platform.  

Navigating Regulatory and Safety Challenges  

Though the promise of drone delivery exists, many complex regulatory and safety issues remain challenges for large-scale operations. To ensure a successful, larger-scale deployment, airspace management, collision avoidance, and compliance with aviation regulations must be given critical consideration. Amazon must partner with regulatory authorities and work closely to develop systems that meet safety standards while also ensuring compatibility with the air traffic control systems currently in place.  

For these reasons, they have begun investing in technologies that enable obstacle detection and avoidance, maintain safe separation distances from persons and buildings, and operate reliably under various environmental conditions. Addressing all of these issues is necessary to build public confidence and ultimately achieve widespread use of their services.  

Environmental and Efficiency Benefits  

Drones can be a great way to reduce logistics-related pollution by using less fuel during delivery when they are electrically powered. Drones produce fewer emissions and perform well over shorter distances, especially in less densely populated suburban areas.   

Creating better drone delivery routes and reducing road-based trips by delivery vehicles can make Amazon’s drone delivery network more sustainable. For the company to scale the technology effectively, it needs to carefully assess energy consumption, battery life, and the equipment’s lifecycle to make the best environmental decisions.  

Competitive Landscape in Delivery Innovation  

Amazon is expanding its drone deliveries amid heightened AI, which has become a major focus for tech retail companies trying to become more efficient in delivering products to the end customers.  

To maintain its dominant position in online shopping & influence other commentators to adopt new technologies, Amazon is spending heavily on developing and deploying drone technology to deliver packages directly from its distribution centers to consumers. If the drone program is successful, it could lead to large numbers of these companies adopting similar technologies, accelerating innovation in last-mile delivery.  

Economic and Workforce Implications  

The growth of drone deliveries means there are many additional economic impacts on the logistics labour force. Automation provides efficiency and cost reductions, but employment in this area will continue to change as drones are used for delivery.  

As a result, potential new jobs may arise in fields related to drone upkeep, drone network management, and artificial intelligence; conversely, delivery positions could either develop or be significantly diminished. Amazon will try to address both the automation of logistics and provide opportunities for a workforce to adapt by developing sustainable, long-term economic benefits from the technology.  

Future of Urban and Suburban Logistics  

Drone delivery networks are likely to change the logistics environment, whether in city or suburban areas. Drone deliveries will initially be deployed in areas with lower population density; however, as navigation, safety, and regulatory advancements continue to develop, more cities will adopt them.  

Drones, when used in conjunction with other delivery technologies such as self-driving vehicles and smart warehouses, will create a 100% automated logistics ecosystem. This interconnected logistics system will provide more efficient, timely delivery for consumers and meet the high demands of online shopping.  

Looking Ahead: A New Delivery Paradigm  

With an investment totalling $5 billion into developing drone delivery systems, Amazon is clearly committed to changing how products are delivered to consumers around the world. By using new technologies such as artificial intelligence (AI), automation, and advanced logistics infrastructure to deliver packages, Amazon is paving the way for a new era in which fast, efficient deliveries will no longer be rare but expected by many consumers.  

As drone technology improves and regulations for safe flying continue to develop, drone delivery could become a regular practice that revolutionises the consumer experience and sets new industry standards. 

A Faster Future for E-Commerce  

The Amazon drone delivery programme is an innovative way to change the face of e-commerce logistics. The company’s goal is to provide very short delivery times and an expandable infrastructure. By doing so, Amazon is making strides in remaining at the cutting edge of the logistics industry.  

If the programme is successful, the network will enable consumers to get their goods even faster than before, fundamentally changing the way they shop. As drone delivery continues to attract investment and improve through technological advances, it may soon become a defining characteristic of how we conduct commerce today.

Source: https://www.amazon.science/blog

FDA Clears NEXUS System for Minimally Invasive Heart Repair in the US

The US Food and Drug Administration (FDA) has approved the NEXUS Aortic Arch Stent Graft System from Endospan for minimally invasive treatment of thoracic aortic arch disease, including chronic aortic dissections. It is intended for patients with dilative lesions in or near the aortic arch who face high risks with traditional open chest surgery.  

Key Points about the NEXUS System Approval: 

  • Approval and indication: The FDA approved the NEXUS system for high-risk patients with aortic arch disease, offering a less invasive alternative to traditional open-heart surgery.  
  • Study data (TRIOMPHE): approval was based on one-year results from the TRIOMPHE investigational device exemption (IDE) study. For patients, this means a 90% survival rate from lesion-related death, 90% were free from disabling stroke, and 98% did not require additional procedures to address leaks, all within one year of treatment.  

Technical Highlights 

  • Bimodular design: features two modules engineered to conform precisely to the ascending aorta and aortic arch anatomy.  
  • Low profile delivery: employs a 20F delivery system currently the lowest in its class  to minimize vascular access complications.  
  • Pre-shaped catheter column enables single-pass access to the arch, minimizing device manipulation and reducing stroke risk.  
  • Integrated branch: contains a dedicated branch for the brachiocephalic trunk to preserve blood flow to critical upper-body vessels.  

Artivion Inc. announced that, following FDA approval, it now has the option to acquire Endospan within 90 days of the clearance date.  

The NEXUS system has already been available in Europe (CE mark) and now provides patients in the US with a specialized, off-the-shelf endovascular solution that offers a minimally invasive alternative to open surgery. This advancement gives patients who were previously limited to open procedures access to a potentially safer, faster recovery option.  

With FDA approval, Artivion may acquire an endorsed plan within 90 days and has arranged financing if it chooses to proceed with the company. The company is reviewing its acquisition option following this approval, which was earlier than expected.  

NEXUS is a branched endovascular stent graft system approved in the US for the minimally invasive treatment of aortic arch disease. Approval was based on the TRIOMPHE study, which showed strong survival, stroke-free rates, and low reintervention rates, as summarized above.  

NEXUS receiving FDA approval ahead of our expectations is an exciting milestone for patients with aortic arch disease, for our partner Endospan, and for Artivion, said Pat McKin, Chairman, President, and Chief Executive Officer. The TRIOMPHE data have continually demonstrated the clinical value of this technology, and we are proud to have supported Endospan on this journey. We have proactively put the financing in place to support a potential acquisition, and we are moving carefully to finalize our considerations around the options. We look forward to providing an update to shareholders soon.  

About Aritvion Inc.  

Artivion Inc. is a medical device company based in suburban Atlanta, Georgia. Its main products include aortic stent grafts, surgical sealants, mechanical heart valves, and implantable cardiac and vascular human tissues. Artivion sells its products in over 100 countries. For more information, visit www. Artivion.com.  

About Endospan Ltd.  

Endospan is a privately held company based in Herzilia, near Tel Aviv, Israel, and a pioneer in endovascular repair of aortic arch disease, including aneurysms and dissections. Endospan received CE mark approval to market the NEXUS stent graft system in Europe, the first half-shelf endovascular system for treating aortic arch disease. This system offers hope to patients with dilative lesions in or near the aortic arch who have been underserved, providing a minimally invasive repair that can lower risk, shorten hospital stays, and speed recovery. While minimally invasive endovascular repair is standard for abdominal aortic aneurysm, patients with aortic arch disease have often only had open chest surgery as an option, which comes with more risks, longer hospital stays, and longer recovery times. For more information about Endospan, visit www.endospan.com.  

Source: FDA Approves Nexus Aortic Arch System  

Tesla Releases Optimus Gen 3 Training for Real-World Home Tasks

Tesla has begun public data collection and training for its third-generation Optimus humanoid robot. This model now focuses on domestic environments rather than factory settings. This change marks a significant milestone in mapping household layouts and object interactions. Tesla is deploying a specialized fleet of learning units to employee residences. By emphasizing advanced spatial reasoning and haptic sensitivity, Tesla aims to transform the robot from an industrial tool into a capable domestic assistant.  

The Engineering of Domestic Dexterity 

The Gen 3 hardware includes redesigned hands with 22 degrees of freedom and integrated tactile sensors. These actuators enable the robot to handle delicate items such as glassware or produce without excessive force. Tactile feedback loops allow real-time perception of texture and weight. Such sensory input is fundamental for activities like folding laundry or clearing a table. In these cases, the grip must be continuously adjusted. Significantly improved environmental sensing now benefits from the vision-depth integration suite in the robots’ heads, which uses eight high-definition cameras. The system creates a 3D map of a home. As a result, the robot navigates around furniture and pets instead of relying on pre-mapped paths; it interprets its surroundings dynamically. Thanks to these abilities, the robot operates safely in cluttered spaces such as kitchens or play areas where obstacles frequently change.  

Training Via Neural Imitation 

A key feature of this release is the human-to-robot imitation training model. Tesla engineers and testers use motion-capture suits to perform common tasks while the robot observes and records movement data. This approach teaches the robot the natural flow of tasks, such as the circular motion required to wipe a counter, by capturing thousands of movement variables. The system builds a generalized understanding of each objective.  

The collected data is uploaded to a central behavioral library. It is then refined through millions of simulated repetitions in virtual environments. The robot practices tasks under different lighting and gravitational conditions. This synthetic hardening ensures the robot can adapt to changes in object placement. For example, it identifies a coffee mug whether it is upright on a coaster or lying in a sink.  

Safety Procedures And Privacy Architecture 

Optimus Gen 3 is designed with safety as its top priority for home use. It features soft-touch joint limiters and a lightweight frame. The robot gives way if it bumps into someone. Its sensors detect people or pets nearby. If this happens, the robot switches to static safe mode immediately and stops any vigorous movements. This approach keeps the robot safe and non-threatening while it learns.  

To protect privacy, all visual data is processed internally on the robot. Its cameras do not record or transmit raw data beyond the device. This ensures household privacy. The robot’s processor converts visualized spatial data. Only these distilled data and intent signals are transmitted to the central system, benefiting all robots.  

Expanding the Domestic Utility Suite 

The first software for Gen 3 focuses on four main jobs: organizing, cleaning, handling packages, and controlling waste in organizing mode. The robot finds misplaced items, such as shoes or remote controls, and returns them to their rightful places for cleaning. It uses regular tones for vacuuming or dusting, maintaining its valence throughout.  

For package handling, the robot retrieves deliveries from the porch and brings them inside. This streamlines online shopping. Its wet sensors assess if a box is fragile or requires special handling. For waste management, it sorts recyclables and moves bins to the curb on schedule. These fundamental tasks will enable even more complex chores in future updates.  

The Future of the Autonomous Household 

As training advances, Tesla will introduce voice-to-action integration. Users will be able to give complex commands, such as “cleanup the kitchen after dinner.” The robot will break these commands into multiple subtasks. It can load the dishwasher and wipe the table in a single workflow. This self-governance requires the system to monitor its power. The robot will return to a docking station when the battery reaches a set level. The goal is a truly set-and-forget appliance that runs effortlessly in daily life.  

The Silent Evolution Of The Home 

We are entering an era where daily life is quietly transformed by advanced automation. Homes grow more responsive as technology adapts to our needs. Chores shift from human hands to automated machines. Over time, the line between tool and inhabitant may blur. Automation manages routine tasks, creating effortless comfort. We trust that our comfort and well-being are reliably supported. The home evolves into a dynamic, responsive space, always ready to serve.

Source: Blog Optimus Tesla