Edition: April 16, 2026

Factory, a startup developing AI agents for enterprise engineering teams, has raised $150 million at a $1.5 billion valuation to build AI coding for enterprises. The company's key differentiator is its ability to switch between different foundation models, and its customers include engineering teams at Morgan Stanley and Palo Alto Networks. Factory aims to compete with other companies vying for dominance in the AI-assisted coding market.

A design flaw in Anthropic's Model Context Protocol (MCP) puts up to 200,000 servers at risk of complete takeover, according to security researchers, who claim Anthropic has declined to patch the issue. The vulnerability affects multiple open-source tools and AI agents, including LangFlow and GPT Researcher. Despite repeated warnings, Anthropic has not modified the protocol's architecture, citing the behavior as 'expected'.

Luma, an AI video generation startup, has launched Innovative Dreams, a production company in partnership with Wonder Project, a faith-focused streaming service, to create religious films and TV shows using AI-powered tools. Their first show, 'The Old Stories: Moses', is set to launch on Prime Video this spring. The partnership aims to leverage AI for faster, cheaper, and more efficient filmmaking without sacrificing quality.

Mozilla has launched Thunderbolt, an open-source AI client that provides data privacy guarantees and an alternative to proprietary enterprise AI platforms from companies like OpenAI and Microsoft. The client can connect to deepset's Haystack platform and allows businesses to self-host their AI workflows. This move aims to give organizations sovereignty and control over their AI operations.

Upscale AI, an AI infrastructure company, is in talks to raise $180-200 million in funding, valuing the company at around $2 billion. This would be the company's third funding round since its launch seven months ago. Despite not having released a product yet, Upscale AI is betting on a full-stack solution for scalable AI infrastructure.

Physical Intelligence, a robotics startup, has developed a new robot brain called π0.7 that can perform tasks it was never explicitly trained on, a significant step towards a general-purpose robot brain. The model can combine skills learned in different contexts to solve new problems, and can be coached through tasks in plain language. This breakthrough could lead to robots being deployed in new environments and improved in real time without additional data collection or model retraining.

OpenAI has revamped its Codex tool with new features, including the ability to operate in the background and control desktop apps, in a bid to compete with Anthropic's Claude Code. The updates enable Codex to perform tasks autonomously, such as iterating on frontend changes and testing apps. This move aims to make Codex a more versatile tool for corporate workflows.

Maine has become the first US state to pass a statewide moratorium on new datacenter approvals, citing concerns over noise, power consumption, and environmental impact. The move is part of a growing trend of local opposition to datacenter development, with several cities and states enacting similar moratoriums. The moratorium aims to give the state government time to assess the impact of these facilities on residents and the environment.

A lawsuit between Elon Musk and OpenAI's Sam Altman is set to go to trial, with Musk accusing OpenAI of straying from its founding nonprofit mission to benefit humanity. The case could impact OpenAI's plans for an IPO and influence how the company controls and distributes its AI technology. The outcome may also affect the future of AI development and safety.

The UK government has launched a $675 million venture fund, Sovereign AI, to invest in domestic AI startups and reduce dependence on foreign technology. The fund will provide access to supercomputers, free visas, and procurement opportunities to portfolio startups. This initiative aims to position the UK as a key player in the AI sector and capture a larger share of the global AI market.

Google has introduced a new AI Mode in Chrome, allowing users to explore the web more seamlessly without switching tabs, and providing a more fluid way to discover and learn from online content. The update enables users to open webpages side-by-side with AI Mode and access powerful tools like Canvas and image creation. This new feature is now available in the US and will expand to other countries soon.

OpenAI has updated its Codex system with new features, including the ability to control desktop apps and generate images, in a bid to compete with Anthropic's Claude Code. The update allows Codex to operate in the background and work in parallel with other agents. This move intensifies the rivalry between OpenAI and Anthropic in the AI development space.

Google has introduced a new feature to its AI Mode, allowing users to explore the web side-by-side with the conversational search experience, making it easier to compare details and ask follow-up questions. This update is available on Chrome desktop and enables users to open web pages alongside AI Mode. The goal is to provide a more seamless and contextual search experience.

Google's latest Chrome update introduces a new AI Mode feature that allows users to open links within the same tab, eliminating the need for tab hopping. The feature also enables searching across multiple open tabs and attaching files and images to queries. This update aims to make search journeys more efficient and context-filled, while potentially increasing user interaction with Google's AI tools.

InsightFinder, a startup using machine learning to monitor and fix IT infrastructure issues, has raised $15 million to help companies diagnose and resolve problems with AI models. The company's solution can detect, diagnose, and prevent issues with AI models by analyzing data, models, and infrastructure together. InsightFinder aims to provide end-to-end feedback loop support for AI observability across development, evaluation, and production stages.

AI traffic to US retailers' websites has risen 393% in the first quarter of 2026, driving higher revenue and conversion rates, with AI visitors engaging more and spending longer on sites. According to Adobe, AI-driven revenue per visit was 37% higher than non-AI traffic in March. Retailers are incentivized to make their sites AI-friendly to capitalize on this trend.

Google's Gemini can now generate personalized images using data from Google Photos, creating tailored responses based on individual tastes and lifestyle. This feature uses prompts to design custom images, such as a dream house or desert island essentials. The integration utilizes labels in Google Photos to identify people and objects relevant to the user.

The Gemini app now uses Personal Intelligence to create personalized images based on users' Google Photos and preferences, eliminating the need for long prompts. Users can generate custom images featuring themselves or loved ones with simple prompts. The feature is rolling out to US subscribers of Google AI Plus, Pro, or Ultra over the next few days.

Roblox is introducing new AI-powered tools to its game development platform, including a 'Planning Mode' that helps creators plan, build, and test games. The tools, which include Mesh Generation and Procedural Model Generation, aim to speed up development and improve collaboration between humans and AI. This update enables developers to create games more efficiently and effectively.

Anthropic has released Claude Opus 4.7, its most powerful generally available model to date, which enhances advanced software engineering tasks and improves image analysis and creativity. This release follows the announcement of Mythos Preview, a cybersecurity-focused model. Opus 4.7 is a step up from its predecessor, Opus 4.6, in complex coding areas.

Google blocked a record 8.3 billion ads globally in 2025, up from 5.1 billion the previous year, with its AI-driven systems catching over 99% of policy-violating ads before they were shown to users. The company attributed the increase to its growing use of AI, particularly its Gemini models, which allow for earlier and more precise detection of problematic ads. This shift reflects a broader change in enforcement, with a growing emphasis on blocking individual ads rather than suspending advertiser accounts.

The UK government faces a potential public backlash against AI unless it can demonstrate that ordinary people will benefit from its widespread adoption, with warnings that failing to address concerns could lead to voter backlash. The Institute for Public Policy Research warns that AI is perceived as a major global risk, alongside climate change and war. Governments must balance protecting people from AI risks with steering transformation to deliver public value.

Runway CEO Cristóbal Valenzuela suggests that Hollywood studios could produce 50 films instead of one $100 million blockbuster by utilizing AI, increasing output and chances of success. This approach could revolutionize the film industry by making it more efficient and cost-effective. Valenzuela believes AI can help the creative class work better and faster, despite controversy and skepticism around its use in creative markets.

Investigative journalist Ronan Farrow discusses his recent article on OpenAI CEO Sam Altman, exploring Altman's trustworthiness and the rise of OpenAI. Farrow's article delves into Altman's tendency to lie and his conflicted persona. The conversation sheds light on Altman's character and the implications of his leadership in the AI industry.

Canva has released its AI 2.0 update, introducing prompt-based editing tools that allow users to create and adjust content by describing their desired outcome to the platform's AI assistant. The update includes a new orchestration layer for Canva's AI models, providing a unified conversational interface. This update aims to make Canva a centralized hub for AI-powered content creation.

Enterprise AI is shifting from a model-centric approach to an operating layer that combines workflow software, data capture, and governance, allowing companies to embed intelligence directly into their operational platforms. Incumbent organizations have an advantage in this space due to their existing proprietary operational data, workforce of domain experts, and accumulated tacit knowledge. This approach enables companies to treat AI as a compounding asset that improves over time.

Canva's updated AI assistant can create editable designs using text prompts, calling various tools to make designs for users, and also integrates with other apps like Slack and Gmail. This update aims to make the AI assistant central to users' workflows, automating content creation and media asset production. The company plans to launch Canva AI 2.0 in research preview this week.

Security researchers have found that Anthropic's AI-powered code reviewer, Claude, can be tricked into approving malicious code by spoofing a trusted developer's identity using forged Git metadata. This vulnerability arises when Claude treats metadata as a signal of trust, rather than independently assessing the code's soundness. The issue highlights the risks of relying solely on AI-powered workflow tools to auto-review and approve pull requests.

Antioch, a simulation startup, has raised $8.5 million to develop tools for robot developers to bridge the gap between virtual and real-world environments. The company aims to provide a scalable solution for training robots, reducing the need for physical testing arenas. With its platform, Antioch wants to make simulation feel like the real world for autonomous systems.

The use of AI in warfare has become increasingly prevalent, with autonomous systems generating targets and controlling drone strikes, but the notion of 'humans in the loop' may be an illusion due to the opaque nature of AI decision-making. AI systems are 'black boxes' that even their creators cannot fully interpret, making it difficult for human overseers to understand their intentions. This lack of transparency raises concerns about accountability and the potential for unintended consequences, including war crimes.

OpenAI has introduced sandbox execution to its Agents SDK, allowing enterprise governance teams to deploy automated workflows with controlled risk and improving the reliability of tasks that require coordination across diverse systems. The updated infrastructure aligns execution with the natural operating pattern of the underlying models, and provides a model-native harness and native sandbox execution. This enables developers to standardize infrastructure and focus on building domain-specific logic that benefits the business.

A Silicon Valley startup, Sabi, is developing a brain-reading beanie that can decode a person's internal speech into words on a computer screen, allowing users to type just by thinking. The non-invasive, wearable device uses EEG and AI to decode neural activity, with an initial typing speed goal of 30 words per minute. The technology has the potential to make brain-computer interfaces accessible to a wide range of people.

Cadence Design Systems has announced partnerships with Nvidia and Google Cloud to expand its AI and robotic capabilities, focusing on combining AI with physics-based simulation and accelerated computing for system-level design. The collaborations aim to improve modelling and deployment in semiconductors and large-scale AI infrastructure. Cadence is also introducing a new AI agent to automate chip design tasks on Google Cloud.

A new approach to measuring human relationships with nature is being developed, focusing on metrics that encourage hope and cooperation rather than dread and separation. The Nature Relationship Index, set to debut later this year, aims to assess how well countries live in harmony with the environment. This shift in perspective could challenge the notion that humans are inherently destructive to nature.

Microsoft, the largest buyer of carbon removal contracts, has paused its purchases, sparking concerns about the future of the carbon removal market. The company has been the primary driver of the market, purchasing around 80% of all contracted carbon removal. Microsoft has stated that it is not permanently ending its carbon removal purchases, but the pause has raised questions about the economics and sustainability of the industry.

The COVID-19 pandemic brought an unexpected benefit to the environment: a significant reduction in noise pollution, allowing animals like the white-crowned sparrow to communicate more effectively. Researchers found that the birds' songs transformed, becoming quieter and more complex, with a richer range of frequencies. This discovery highlights the need to address anthropogenic noise as a form of pollution.

DeepL, a renowned text translation company, has launched a voice-to-voice translation suite for various use cases, including meetings and group conversations, with plans to develop an end-to-end voice translation model. The company is also releasing an API for customized use cases and add-ons for platforms like Zoom and Microsoft Teams. This innovation aims to improve real-time voice translation and provide more accurate results.

Researchers have introduced Ecom-RLVE, a framework for training e-commerce conversational agents using reinforcement learning with verifiable rewards. The framework provides eight adaptive verifiable environments that simulate real-world shopping scenarios, allowing agents to learn task completion skills. Early results demonstrate the effectiveness of environment scaling and adaptive difficulty in achieving agentic, real-world task completion.

Anthropic introduces Claude Opus 4.7, a significant improvement over its predecessor Opus 4.6, with enhanced capabilities in advanced software engineering, vision, and cybersecurity safeguards. The model is now generally available across all Claude products and major cloud platforms. Opus 4.7 offers improved performance, precision, and creativity, making it a game-changer for developers and cybersecurity professionals.

The rise of code agents has led to an influx of low-quality pull requests in open-source projects, forcing maintainers to rethink their approach to open-source contributions. To address this issue, a Skill and test harness have been developed to help port language models from transformers to mlx-lm. This aims to support contributors and reviewers in creating high-quality model ports.

A new blog post demonstrates how to train and finetune multimodal embedding and reranker models using the Sentence Transformers library, achieving improved performance in visual document retrieval tasks. The post provides a practical example of finetuning a model for visual document retrieval, resulting in a significant gain in performance. This has potential applications in retrieval augmented generation, semantic search, and more.

Mike Krieger, Anthropic's chief product officer, has resigned from Figma's board after reports emerged that Anthropic's next model will include design tools competing with Figma. This move has sparked concerns about the potential dominance of AI labs in the software industry. Figma's stock price has risen 5% since Krieger's departure was disclosed.

Anthropic is expanding its presence in London with a new 158,000-square-foot office, aiming to increase its research and commercial footprint in Europe and compete with other leading AI labs for top talent. The move follows a reported attempt by UK government officials to coax the company into expanding in London after a fallout with the US administration. The expansion will also deepen Anthropic's work with the UK's AI Security Institute.

Cyberscammers are using illicit tools sold on Telegram to bypass banks' security measures, while Microsoft's pause on carbon removal purchases has sparked concerns about the future of the industry. A new measurement index is also being developed to understand human relationships with nature.

Anthropic has revised its pricing plan for enterprise customers, shifting them from a seat-based model to a usage-based plan, which may increase costs for large organizations. The change comes as the company faces turmoil and prepares for a potential IPO. This move is expected to push enterprises towards metered pricing.

Anthropic, the developer of AI assistant Claude, may require users to verify their identity to access certain features, using a third-party vendor called Persona that has sparked controversy in the past. The move aims to prevent abuse and comply with legal obligations, but has raised concerns among users about data privacy. Anthropic claims it will only collect minimal information and not use it to train its models.

Google has updated its AI Mode in Chrome, allowing users to open links to sources alongside their chat without leaving the page. This new feature enables users to ask follow-up questions about the content on the page. The update is part of Google's ongoing efforts to enhance its chatbot-like search feature.

Character.AI has introduced a new 'Books' mode, allowing users to engage in roleplaying experiences within familiar worlds from classic public domain titles. This move aims to shift perceptions of AI roleplay beyond controversy. The feature launches with over 20 titles, including Alice in Wonderland and Pride and Prejudice.

Meta is increasing the prices of its Quest 3 and Quest 3S virtual reality headsets due to a global RAM shortage, with price hikes ranging from $50 to $100. The new prices will take effect on April 19 and also apply to refurbished units. This move follows similar price increases by other tech companies like Samsung, Microsoft, and Sony.

Public sector organizations face unique constraints in adopting AI due to security, governance, and operational requirements, but purpose-built small language models (SLMs) offer a promising solution. SLMs provide greater security and control by being housed locally and using fewer parameters, making them more suitable for government institutions. This approach can help overcome the challenges of AI deployment in the public sector.

The article highlights the top cloud migration software for Infrastructure as Code (IaC) deployment, with Infros standing out for its cloud architecture design and validation capabilities. Other notable tools include Spacelift, which excels in IaC orchestration. These tools help teams migrate workloads to the cloud while ensuring reproducibility and deployment readiness.