Edition: June 30, 2026

The Trump administration has lifted export controls on Anthropic's powerful AI models, Mythos and Fable, after the company agreed to strengthen safeguards against security risks. The move comes after Anthropic worked with the Commerce Department to address concerns over cybersecurity and jailbreaks. The decision allows for the export and transfer of the models without a license.

OpenClaw, a free and open-source AI agent, is now available as an app on both Android and iOS devices, allowing users to run their agents from their pocket. The app connects to the OpenClaw Gateway, enabling access to various AI tools and skills. This development marks a significant expansion of AI agents into mobile devices.

Anthropic has launched Claude Science, a new product designed to support scientific research, particularly in computational biology and drug development. The product can autonomously carry out meaningful work and has access to specialized tools, making it a significant addition to Anthropic's offerings. Claude Science is now available to all paid Claude subscribers.

Former DeepMind researchers who created an AI that beat humans at poker have applied the same technology to trading stocks, with their company EquiLibre Technologies now valued at $500 million. Their algorithms have been trading billions in daily volume and have a perfect record of zero negative months since inception. EquiLibre's success has made it an attractive investment, with a recent Series A funding round led by Creandum.

A new attack has been discovered that can trick AI browsers into entering a false reality, bypassing their guardrails and allowing for destructive actions. This vulnerability highlights the risks of using AI browsers, which can blur the line between browsing and taking sensitive actions. The attack demonstrates the limitations of current security measures, which are reactive rather than proactive.

Researchers introduce ScarfBench, a benchmarking tool for evaluating AI agents on enterprise Java framework migration tasks, highlighting the challenges of framework migration and the limitations of current AI agents. The benchmark assesses the ability of AI agents to successfully build, deploy, and preserve the behavior of migrated applications. Initial results show that even state-of-the-art agents struggle with framework migration, achieving less than 10% behavioral success.

Etched, a Nvidia competitor, has reached a $5 billion valuation and $1 billion in sales for its AI chip, with the startup booking contract orders for its 'frontier inference clusters' that aim to improve inference speed and efficiency. The company has raised $800 million to date, including a recent $500 million round. Etched's progress marks significant growth in the AI chip market, with investors showing strong interest in the technology.

Anthropic has launched Claude Sonnet 5, a more powerful and agentic version of its midsize model, offering a cheaper alternative to run agents with performance close to that of Opus 4.8. The new model demonstrates significant improvements over its predecessor in agentic performance and safety. It will be the default model for free and Pro plans and is available for every subscription at a lower price point than competitors.

Acti, a Singapore-based startup, has launched an AI-powered keyboard for iOS and Android that can take actions on behalf of users, bringing AI tools directly into apps like email, messaging, and social media. The keyboard, powered by Google's Gemini models, can perform tasks such as local recommendations and live stock prices, and allows users to create custom shortcuts called Skills. This new approach embeds AI into existing interfaces, rather than requiring users to open separate AI chatbots.

Scientists are exploring ways to reverse aging by reprogramming cells to a younger state, with billions of dollars invested in the research. A recent conversation with experts delved into the potential of these experimental treatments and their possible timeline. The approach has gained significant attention, with notable investments from prominent figures.

Anthropic has introduced Claude Science, an AI workbench designed to streamline computational research for scientists by providing a single environment to access databases, pipelines, and tools. The workbench utilizes existing Claude models and focuses on workflow efficiency rather than introducing new models. This launch is part of Anthropic's strategy to provide industry-specific solutions beyond just AI models.

Google DeepMind has released Nano Banana 2 Lite, its fastest and most cost-efficient image model, and Gemini Omni Flash, a high-quality video generation and conversational editing model. These models aim to make it easier for developers to experiment and scale their ideas. They are available in Google AI Studio, Gemini API, and other platforms.

Amazon Web Services has launched a new $1 billion organization for AI-focused forward-deployed engineers, aiming to help companies integrate AI solutions. The organization will embed engineers within companies to deploy purpose-built agents and provide lasting AI skills. This move follows similar initiatives by OpenAI and Anthropic.

A recent paper by Goldfeder, Wyder, LeCun, and Shwartz-Ziv suggests that specialization is inevitable in AI systems due to optimization theory, evolutionary biology, and machine learning principles. The research indicates that systems that achieve significant results are often narrowly focused on a specific domain, rather than being general-purpose. This pattern is consistent across various domains and architectures, implying a common cause that favors specialization over generality.

The agriculture industry is poised to benefit from AI, with potential improvements in crop yield, water use, and chemical usage, but its effectiveness relies on a solid data foundation, which is often lacking. Without clean and accurate data, AI solutions can generate misleading outputs, leading to counterproductive decisions. Industry leaders must prioritize data readiness to unlock the full potential of AI in agriculture.

Senator Bernie Sanders is advocating for firm regulation of the AI industry, proposing a moratorium on data center construction and a sovereign wealth fund to tax AI companies and provide direct payments to citizens. Sanders believes AI is a transformative technology that requires careful governance to benefit ordinary people, not just the wealthy elite. His efforts come as public anger grows over the impact of AI and data centers on communities.

Zurich, Switzerland has emerged as a major R&D hub, attracting tech giants like Google, Apple, and Microsoft, due to its unique combination of access to major markets, political stability, and strong intellectual property protection. The region's high density of AI research and commercialization is supported by its highly developed innovation ecosystem and venture capital investment in deep tech. This has made Zurich an attractive location for companies building specialized AI capabilities.

Crypto exchange OKX has launched a marketplace called OKX AI, where AI agents can hire and pay each other autonomously, settling transactions using blockchain-based payments and stablecoins. The platform aims to become a key infrastructure for the emerging 'agent economy', where AI agents can transact and build reputations. OKX's launch marks a significant push into the fintech space, beyond its core crypto trading business.

A new study reveals that the top 15% of AI users in the UK report stronger performance reviews and pay increases, but the remaining 85% need upskilling to unlock similar benefits. The research identifies uneven AI adoption across the workforce, with most still in early-stage adoption. To address this, a new AI skills quiz has been launched to help individuals benchmark their skills and progress to advanced AI use.

A recent report challenges the notion that AI will lead to broad job losses, finding that companies heavily investing in AI are actually growing their headcount faster, including in entry-level roles. However, the data is skewed towards tech-forward firms, and the benefits of AI may not be universal. The report suggests that AI can be a tool for firm expansion, but may also widen the gap between companies with and without the resources to invest in AI.

Hugging Face Model Pages now feature Every Eval Ever results, enabling cross-posting and interpreting evaluation results while linking to open models and leaderboards. This integration aims to improve the reporting and comparison of AI evaluation results. The unified metadata store now contains over 229,000 evaluation results across 22,000 models and 2,200 benchmarks.

Google has released Nano Banana 2 Lite, a faster and more affordable version of its AI image generator, capable of producing images in four seconds at a cost of $0.034 per 1,000 images. This new model is optimized for high-volume workflows and is now available through Google AI Studio and other platforms. The release aims to support rapid image generation and video creation for various use cases.

Research suggests that referring to AI tools as 'coworkers' can lead to decreased productivity, while companies like Microsoft and Google are releasing tools to manage teams of AI agents. Meanwhile, a solar-powered platform is being developed to deliver better internet from the stratosphere. These advancements are changing the way we work and connect.

Base44, a vibe coding platform, has launched its own AI model to support users in creating apps with natural language, aiming to outperform frontier models and reduce costs. The move is seen as a way for the company to increase its defensibility in a competitive market. This development reflects a broader trend in the AI industry, where companies are seeking to own and optimize their own models.

Google's NotebookLM now generates 60-second TikTok-style AI clips to summarize research, available to Google AI Ultra and Pro subscribers. The feature uses AI art and narration to create engaging videos. This adds to existing features like AI podcasts and cinematic videos.

Netflix's upcoming reality show 'Wonka's The Golden Ticket' will feature an AI-generated voice of Gene Wilder, created with the consent of his family. The show is set to premiere on September 23rd and is part of Netflix's partnership with the Roald Dahl company. The AI voice was developed by ElevenLabs, a company that has also worked on re-creating the voices of Michael Caine and Stan Lee.

Libby, a popular ebook lending app, is introducing AI content controls to help readers filter out AI-generated books. The move comes as the digital publishing industry prepares for a wave of AI-generated content. Libby's new feature will allow readers to select whether they want to see AI-generated books in their search results.

X has introduced a hosted Model Context Protocol (MCP) server, making it easier for AI tools to connect to the platform using a user's account permissions. This move simplifies the integration process for developers and positions X as an information network with real-time data. The hosted MCP server does not enable autonomous posting and is subject to X's API rules to prevent spam.

Riverside, a podcasting platform, is expanding its services to include newsletter publishing, allowing users to turn their existing videos and podcasts into newsletters using AI. The company aims to provide a seamless experience for creators, leveraging their existing content to reach audiences. Riverside's move is part of a larger trend of platforms diversifying into alternative publishing avenues.

Proton's privacy-focused AI chatbot, Lumo, has been upgraded to version 2.0, featuring new image recognition and generation capabilities, as well as improved performance and user-controlled persistent memory. The update also introduces a 'thinking mode' for complex problems and maintains Proton's zero-access encryption architecture for enhanced user privacy. Lumo 2.0 is available now, with free and paid tiers offering varying levels of access and resources.

Microsoft has developed a new feature to prevent unwanted bots from joining Teams meetings, requiring human approval for bot participation. The technology uses behavioral and infrastructure signals to identify bots, and will eventually allow registered bot developers to include a self-identification marker. This aims to strengthen Teams' security and privacy by keeping sensitive discussions bot-free.

Bill Savitt, a lawyer, has successfully represented clients against Elon Musk in two cases, including Musk v. Altman, where Musk sued Sam Altman and OpenAI. Savitt's mild-mannered and soft-spoken cross-examination style has been noted. Musk has accused Savitt of asking unfair questions during the trial.