Edition: April 20, 2026

Vibe coding platform Lovable has denied a data leak, citing 'intentional behavior' and blaming bug-bounty service HackerOne, despite a researcher finding that anyone could access sensitive user info. The leak stems from a Broken Object Level Authorization vulnerability, which Lovable claims is by design for public projects. The company's response has been criticized as an example of an AI firm shirking responsibility for security flaws.

Anthropic has secured a $5 billion investment from Amazon, bringing the total investment to $13 billion, in exchange for a commitment to spend $100 billion on Amazon Web Services over the next decade. The deal involves the use of Amazon's custom chips, including the Trainium AI accelerator chip. This partnership echoes a similar agreement between Amazon and OpenAI.

A humanoid robot developed by Chinese company Honor has set a new half-marathon record, completing the 13.1-mile race in 50 minutes and 26 seconds, beating the human record by seven minutes. The robot, which was designed to mimic the human body, incorporated advanced features such as long legs and a liquid cooling mechanism. This achievement marks a significant milestone in the development of humanoid robots and raises questions about the potential for machines to outperform humans in physical tasks.

Claude Desktop, an AI app from Anthropic, has been found to install files that affect other browsers without user consent, even before those browsers are installed, raising concerns over privacy and security. This practice has been labeled as 'spyware' by a privacy consultant, who claims it breaches European privacy law. The app's behavior allows it to pre-authorize browser extensions without permission, potentially creating security risks.

The phrase 'it's not just X, it's Y' has become increasingly common in corporate communications, with usage more than quadrupling from 2023 to 2025, suggesting a growing reliance on AI-generated writing. This trend is seen in reports from companies such as Cisco, Accenture, and Microsoft. The phrase is now considered a potential indicator of AI-generated text.

Epic Games has introduced a new 'conversations' tool, allowing Fortnite creators to develop AI-powered characters that can engage in unscripted dialogue with players. This feature enables NPCs to think, know, and behave based on simple prompts and selected voices. The tool opens up new possibilities for interactive storytelling in the game.

A recent study found that making AI chatbots seem nicer and more friendly can make humans interact with them as if they were human, even if the underlying system hasn't improved. The study analyzed over 2,000 human-LLM interactions and found that warmth and empathy can significantly impact user perceptions. However, excessive friendliness can also lead to overtrust and susceptibility to deception or manipulation.

The National Security Agency is reportedly using Anthropic's Mythos model, despite the Pentagon labeling Anthropic a 'supply-chain risk' due to restricted access to its full capabilities. The NSA is using Mythos to scan for vulnerabilities, while the US military argues in court that Anthropic's tools can threaten national security. This development comes as Anthropic's relationship with the US administration appears to be improving.

Google has introduced a new Android command-line interface (CLI) designed for AI agents, aiming to reduce token usage by 70% and task completion time by three times. The CLI is not a replacement for Android Studio, but rather a tool for agents working outside the IDE. It enables agents to build Android applications more efficiently, with features like creating applications from templates and managing the Android SDK.

Britain's datacenter map is shifting away from London due to power shortages, planning constraints, and reduced reliance on low-latency connections, with experts suggesting other locations may become more attractive for AI datacenter capacity. The UK government's AI Opportunities Action Plan aims to drive economic recovery by positioning the country at the forefront of AI development. This shift is driven by the growing power demands of AI infrastructure and the need for more efficient and sustainable datacenter operations.

Huawei's HiFloat4 training format has been found to be superior to Western-developed MXFP4 in a bakeoff on Ascend chips, highlighting Chinese interest in maximizing training and inference efficiency. Meanwhile, Anthropic has made early progress in automating AI safety research and development. These advancements have significant implications for the future of AI research and development.

The UK government has launched an £80 million AI procurement initiative, part of its £500 million sovereign capability fund, to help validate new AI capabilities and de-risk investment for the wider market. The government plans to set challenges in areas such as scientific discovery, health, and national security, with contracts worth up to £5 million per project. Successful bidders will retain ownership of the intellectual property developed during the project.

Scientists are warning that 'mirror' bacteria, a synthetic form of life, could potentially threaten all life on Earth, while Chinese tech workers are pushing back against the use of AI doubles that could replace them. These developments highlight the risks and challenges associated with emerging technologies.

A recent ISACA report reveals that most organizations are unprepared to handle AI system incidents, with 59% of digital trust professionals unable to explain how quickly they could stop an AI system emergency. This lack of preparedness poses a significant risk of irreversible damage. Experts emphasize the need for effective AI governance and management to mitigate these risks.

Tech CEOs like Mark Zuckerberg and Jack Dorsey are exploring the use of AI to create digital avatars and streamline management hierarchies, aiming to exert greater influence within their companies through AI-enabled omnipresence. Zuckerberg is developing a photorealistic AI avatar to interact with Meta staff, while Dorsey envisions a company where all employees report directly to him through an AI layer. This trend suggests that top executives are determined to leverage AI to reshape their roles and companies.

Colossal Biosciences claims to have cloned four red wolves, a species declared extinct in the wild in 1980, sparking surprise and debate among scientists and conservationists. The clones are said to be part of the company's de-extinction efforts, which also include the cloning of dire wolves. The authenticity and purpose of the cloning project have raised questions within the scientific community.

The UK's Science, Innovation and Technology Committee has launched an inquiry into low-energy computing to curb datacenter energy use driven by AI, exploring emerging technologies like neuromorphic computing and silicon photonics. The inquiry aims to assess the maturity and potential of these technologies to reduce energy consumption. The UK's datacenter energy demand is expected to quadruple by 2030, posing a challenge to the country's net-zero targets.

Chinese tech workers are being encouraged by their bosses to train AI agents to replace them, prompting a wave of soul-searching among employees. A viral GitHub project called Colleague Skill has sparked debate about workers' dignity and individuality in the age of AI. The project allows users to 'distill' their colleagues' skills and personality traits and replicate them with an AI agent.

The AI industry's reliance on token-based pricing models is flawed and may lead to 'token incremental burn syndrome' (TIBS), where costs inflate without a clear understanding of the value provided. This model can result in inefficiencies and lock-in for subscribers. The lack of alternative metrics makes it difficult to link costs to actual value.

Most AI initiatives, particularly generative AI, are being canceled at the pilot stage due to a lack of measurable returns, with around 95% of projects failing to deliver. A recent MIT report highlights the shocking truth behind the high failure rate of enterprise AI projects. Successful initiatives have one thing in common, setting them apart from the rest.

Anthropic CEO Dario Amodei met with White House officials to discuss the company's AI cybersecurity tool, Mythos, which has impressed agencies with its ability to identify vulnerabilities. The meeting marks a reversal of the Trump administration's previous stance on Anthropic, which had been deemed a supply chain risk. The US government is now interested in utilizing Mythos to strengthen its cyber defenses.

Vercel, the company behind the Next.js web development framework, has reported a security incident that led to the compromise of some customer credentials, blaming a third-party AI tool called Context.ai for the breach. The incident occurred when an attacker gained access to a Vercel employee's Google Workspace account through Context.ai. Vercel has notified affected customers and deployed protection measures.

Microsoft's GitHub has stopped accepting new Copilot account sign-ups due to a capacity crunch, citing increased compute demands from agentic workflows and long-running sessions. The pause aims to help the company serve existing customers more effectively. GitHub will also tighten usage limits for individual plans to balance reliability and demand.

Bobyard 2.0, an AI platform for construction and landscaping estimators, has been launched with improved takeoff workflows and a unified AI workbench to reduce errors and increase efficiency. The update features a 'measure first, price later' model, Multi-Measure, and an AI Workbench with Review Workflow. Bobyard 2.0 aims to streamline the estimating process, reducing takeoff times and improving accuracy.

Google has expanded its Gemini in Chrome feature to seven new countries, including Australia, Indonesia, and South Korea, allowing users to access AI-powered tools and services directly within the browser. The feature, which includes a sidebar-based assistant and image transformation capabilities, is now available in more countries after initial launches in the US, India, Canada, and New Zealand. Gemini in Chrome aims to enhance user productivity and experience with personalized answers and integrated services.

Fermi, an AI nuclear power firm, sees sudden departures of its CEO and CFO, with shares plummeting 22% on Monday. The company's co-founder Toby Neugebauer steps down as chairman but remains on the board. Lead Independent Board Director Marius Haas takes over as chairman.

A commentary piece argues that Silicon Valley has lost touch with the needs of ordinary people, with tech enthusiasts overly excited about discoveries that may not be as groundbreaking as they seem. The example given is the supposed revelation that language models can understand context and generate human-like text. The author suggests that such discoveries are not as revolutionary as they are made out to be.

Prego, the pasta sauce company, has developed a device called The Connection Keeper that records dinner conversations to encourage families to put away their phones and engage with each other. The device, created in partnership with StoryCorps, can record up to eight hours of audio and allows users to share their recordings on the StoryCorps website. The goal is to promote meaningful conversations and preserve family stories.