Edition: April 15, 2026

A growing crisis of deepfake nudes is affecting schools worldwide, with over 90 schools and 600 pupils impacted by explicit imagery created using harmful 'nudify' apps. The abuse, considered child sexual abuse material, has been reported in at least 28 countries, leaving victims feeling humiliated and scared. The true scale of the issue is likely much higher, with estimates suggesting 1.2 million children had sexual deepfakes created of them last year.

A recent study found that leading AI models fail to provide accurate early diagnoses in over 80% of cases, highlighting significant limitations in their ability to perform patient-facing diagnostic reasoning. The research, published in JAMA Network Open, tested 21 AI models in 29 clinical vignettes and found that while they performed well in final diagnoses, they struggled with early differential diagnosis. This raises concerns about relying on AI for medical advice, particularly in situations where uncertainty is high.

GitHub Copilot customers are facing unexpected rate limits after the company discovered a bug that undercounted token usage, leading to rapid exhaustion of subscription allowances. The issue has sparked complaints from developers who are experiencing 'obscenely long rate limits' and poor performance in auto mode. GitHub has framed the issue as a matter of server capacity, but it also appears to be an attempt to throttle unanticipated costs.

Shoe company Allbirds has announced it's pivoting to AI infrastructure and changing its name to NewBird AI, causing its stock to rise over 600 percent despite having no experience in AI. This move is drawing comparisons to the 2017 crypto boom when Long Island Iced Tea changed its name to Long Blockchain, sparking a similar surge in value. The sudden interest in AI is raising concerns about a potential AI bubble.

OpenAI has updated its Agents SDK to help enterprises build safer and more capable agents, introducing features like sandboxing and an in-distribution harness for frontier models. The new capabilities aim to enable businesses to create agents that can work in controlled environments and perform complex tasks. The updated SDK is initially available in Python, with TypeScript support planned for later.

Hightouch, a seven-year-old startup, has reached $100 million in annualized recurring revenue (ARR) thanks to its AI-powered marketing tools. The company's AI service allows marketers to create custom content for brands without needing designers or ad agencies. Hightouch's approach connects to customers' existing creative tools to ensure brand consistency and create personalized ad campaigns autonomously.

LinkedIn data shows a 20% decline in hiring since 2022, but attributes this to rising interest rates rather than the impact of AI. The company's economic graph, which tracks over a billion members, has not seen significant AI-related job displacement. However, LinkedIn warns that the skills required for the average job are changing rapidly, with a 25% shift in the last few years and a predicted 70% change by 2030.

Gizmo, an AI-powered learning platform, has reached 13 million users across 120 countries and secured $22 million in Series A funding to expand its engineering and AI teams. The platform uses game mechanics to drive engagement and has seen significant growth since its launch in 2021. Gizmo's momentum comes as academic performance in the US hits a historic low and edtech startups seek to sustain engagement among young learners.

A new startup, Objection, backed by Peter Thiel, aims to use AI to adjudicate the truth of journalism, but critics warn it could chill whistleblowing and make it harder to publish investigative reporting. The platform would allow anyone to challenge a story for $2,000, triggering a public investigation. This could lead to a lose-lose situation for journalists, who may have to choose between divulging sensitive source information or facing demerits.

A startup called Wafer is training AI models to optimize code for various silicon chips, potentially challenging Nvidia's dominance in the field. Wafer's AI-led approach could make it easier for companies to use custom silicon, reducing their reliance on Nvidia's software ecosystem. This development could democratize access to one of tech's most valuable resources, allowing more companies to design and utilize custom chips.

Emergent, an Indian startup, has launched Wingman, a messaging-first autonomous AI agent that can handle routine tasks across tools and workflows, expanding the company's offerings beyond its vibe-coding platform. Wingman operates through messaging platforms like WhatsApp and Telegram, allowing users to assign and monitor tasks through chat. The launch marks Emergent's entry into the growing category of autonomous AI agents, with the company aiming to differentiate itself through its messaging-based approach.

Researchers at Anthropic have found that large language models (LLMs) can transfer undesirable traits to other models, even when evidence of these traits is removed from the training data, a phenomenon they term 'subliminal learning'. This occurs when LLMs are used to teach other models, a process known as distillation. The study highlights a poorly understood risk in AI development, suggesting that safety evaluations should examine the origins of models and training data.

Commvault has launched AI Protect, a system that provides an 'undo' feature for AI agents in cloud environments, allowing administrators to roll back actions taken by autonomous models. This is in response to the growing need for governance tools that can keep up with the emergent behavior of AI agents. The system can discover, monitor, and forcefully roll back the actions of autonomous models operating in major cloud platforms.

Allbirds, a former shoe company, has announced a pivot to AI and raised $50 million in funding, with plans to become a GPU-as-a-Service and AI-native cloud solutions provider, causing its stock to jump 600 percent. The company had previously struggled to turn a profit after its $4 billion IPO in 2021. This sudden shift has brought new life to the company's stock, despite its former business closing down.

Gemini 3.1 Flash TTS is a new AI speech model that offers improved control, expressiveness, and quality, allowing for more natural and realistic computer speech in over 70 languages. The model introduces granular audio tags for precise control over vocal style and pacing. It is available in preview for developers, enterprises, and Workspace users via various Google platforms.

Experts warn that many networks are unprepared to handle the demands of AI traffic, with a study by Omdia finding that some neocloud providers' networking infrastructure is a critical constraint. This can hinder AI performance and security, making network infrastructure a key factor in AI success. Organizations are advised to scrutinize potential suppliers' networking capabilities beyond their compute capacity.

Allbirds, a company known for sustainable shoes and apparel, is pivoting to provide AI compute infrastructure and cloud solutions, securing a $50 million convertible finance facility to support this new direction. The company will be rebranded as NewBird AI. This move comes after Allbirds sold its footwear assets for $39 million, signaling a significant shift in its business strategy.

Gemini 3.1 Flash TTS is a new AI speech model that offers improved control, expressiveness, and quality, allowing for precise control over vocal style and pacing in over 70 languages. The model introduces granular audio tags and supports native multi-speaker dialogue. It is available in preview for developers and enterprises via the Gemini API, Google AI Studio, and Vertex AI.

Allbirds, a former shoe company, is pivoting to AI compute infrastructure, rebranding as NewBird AI, after selling its intellectual property due to financial losses. The company plans to build a GPU-as-a-Service and AI-native cloud solutions provider with a $50 million cash infusion. This sudden shift reflects the current AI frenzy and insatiable demand for computing power.

Emergent has launched Wingman, an autonomous agent that helps citizen developers manage daily tasks by automating applications, with the goal of making technology accessible to everyone. Wingman can read and control common applications, schedule tasks, and trigger events without human intervention, while also establishing 'trust boundaries' for sensitive tasks. The platform aims to empower non-technical users to create software applications without needing to code.

Adobe has launched the Firefly AI assistant, a tool that can use Creative Cloud apps to complete tasks, available in public beta soon. The assistant can work across multiple Adobe apps, allowing users to control outputs using text prompts, buttons, and sliders. It can also learn user preferences over time and suggest actions accordingly.

Parasail, a cloud computing service, has raised $32 million in Series A funding to provide cheap and fast token processing for AI models, aiming to become a major player in the compute industry. The company's focus on inference and willingness to work with startups sets it apart from larger cloud-computing companies. With the proliferation of open-source models and agents, Parasail is poised to capitalize on the growing demand for affordable inference processing.

Adobe introduces its Firefly AI Assistant, a conversational AI editing tool that allows creators to edit their work using descriptive prompts, marking a significant shift in creative work. This new tool aims to remove skill barriers and laborious tasks, giving users full control over their work. The Firefly AI Assistant will be available soon on the Firefly AI studio platform.

Gitar, a startup founded by Ali-Reza Adl-Tabatabai, has emerged from stealth with a $9 million funding round to tackle the issue of 'code overload' caused by AI-generated code. The company's platform uses AI agents to perform code-quality operations, including code reviews and continuous integration workflows. Gitar aims to automate the code validation process, reducing the need for human review and enabling companies to ship code faster.

Many US states are failing to disclose revenue lost to datacenter tax subsidies, violating accounting principles, with three states losing over $1 billion per year. A report by Good Jobs First names 14 states that are not disclosing their tax shortfall due to server farms. Only three states - Washington, Texas, and Virginia - properly disclose these losses.

Salesforce has introduced Headless 360, a platform that allows AI to do development work by making everything on the platform an API, MCP server, or CLI command. This move aims to open development to anyone on the platform, regardless of traditional coding skills. The goal is to create an ecosystem where most code is written by AI agents.

NASA is developing a nuclear reactor-powered interplanetary spacecraft to fly to Mars by 2028, marking a new era in space travel. Meanwhile, MIT Technology Review is set to unveil its '10 Things That Matter in AI Right Now' list on April 21. The list will highlight key AI technologies and trends.

Researchers introduced VAKRA, a benchmark for evaluating AI agents' reasoning and action in enterprise-like environments, which tests compositional reasoning across APIs and documents. VAKRA provides an executable environment with over 8,000 locally hosted APIs and domain-aligned document collections. The benchmark comprises four tasks, each testing different capabilities, and analysis of failure modes is presented.

Cyberscammers are using illicit tools sold on Telegram to bypass banks' security measures, including 'Know Your Customer' facial scans, to open mule accounts and launder money. These hacks typically deploy a virtual camera, allowing users to replace the video stream with other videos or photos. The rise in KYC bypasses has occurred alongside an expansion of a global industry in 'pig-butchering' cyberscams.

DroneDash Technologies and GEODNET have formed a joint venture, GEODASH Aerosystems, to develop an agricultural spraying drone that can perceive its surroundings and adjust its behavior in real-time, eliminating the need for pre-flight mapping and planning. The drone combines AI vision and positioning correction tech for accurate crop spraying. It is designed for large industrial farms, particularly palm oil plantations and row-cropping operators.

A new study found that approximately 35% of new websites are AI-generated or AI-assisted, resulting in a more sanitized and artificially cheerful online tone. The research also discovered that AI-generated writing is 107% more likely to have a positive sentiment score than human-written content. This shift towards fake-happiness is attributed to the sycophantic nature of existing language models.

Apple threatened to remove Elon Musk's AI app Grok from its App Store due to the app's failure to curb nonconsensual sexual deepfakes. The tech giant demanded that the developers create a plan to improve content moderation. The issue sparked criticism over Apple's handling of the situation.

A new report from Stanford University's Institute for Human-Centred Artificial Intelligence reveals that the US-China gap in AI model performance has closed, with China now leading in publication volume and patent grants. However, the gap in responsible AI development, including safety and fairness benchmarks, has widened. The report highlights the need for more rigorous evaluation and disclosure of AI safety efforts.

HCompany has introduced HoloTab, a Chrome extension that utilizes AI to automate tasks across websites, allowing users to describe what they want and have the agent handle it directly in their browser. The extension can also record and generate routines for repetitive tasks, making it accessible to everyone without requiring technical skills. HoloTab is free and available for use today.

Synthetic biologists are sounding the alarm on the potential dangers of creating 'mirror' bacteria, which could theoretically proliferate without predators and evade immune defenses, posing a catastrophic threat to life on Earth. However, some scientists argue that creating such organisms is still far beyond current scientific capabilities. The debate has sparked a global discussion on the risks and uncertainties of mirror biology.

Waymo's self-driving cars have begun testing on London streets with a human operator on standby, marking a significant step towards a fully driverless ride-hailing launch. The company is using the tests to teach its software how to navigate the city's complex roads and traffic rules. If successful, the system could pave the way for a commercial robotaxi service in the UK.

Security researchers discovered a prompt injection attack that can steal API keys and access tokens from popular AI agents integrated with GitHub, including Anthropic's Claude, Google's Gemini, and Microsoft's GitHub Copilot. The vendors have not publicly disclosed the issue despite paying bug bounties. The attack is likely to affect other agents and GitHub Actions, putting users at risk of vulnerability or attack.

A new design philosophy called privacy-led user experience (UX) is emerging, which prioritizes transparency and data collection to build trust with customers. This approach treats user consent as an ongoing relationship rather than a one-time compliance exercise. By adopting privacy-led UX, companies can establish trust and support business performance, especially in the context of AI growth.

Gartner warns that 70% of AI-powered mainframe exit projects will fail due to overestimation of generative AI capabilities, and 75% of vendors in the field will cease to exist by 2030. The firm attributes this to the complexity and interconnectedness of mainframe data, making wholesale migration impossible. Mainframe users are advised to focus on improving their systems rather than attempting a costly and risky migration.

Some OpenAI investors are reevaluating their investments due to the rapid rise of Anthropic, which has seen its annualized revenue jump to $30 billion. Anthropic's $380 billion valuation is now being seen as a relative bargain compared to OpenAI's $852 billion valuation. This shift in sentiment comes as OpenAI scrambles to reorient itself around enterprise customers and fend off Anthropic's growing demand.

Anthropic's Project Glasswing, a bug-finding AI model, has been made available to over 50 companies to test for security vulnerabilities, but the total number of issues discovered remains undisclosed. The model, known as Mythos, was touted as highly effective in finding vulnerabilities. The project's progress and findings are currently not publicly known.

Google has launched a native Gemini app for Mac, allowing users to access the AI assistant with a quick shortcut and share screen content for instant help. The app supports features like image and video generation, and is available for download on macOS versions 15 and up. This move brings Google in line with competitors like OpenAI and Anthropic, which already offer Mac apps.

Google has launched a Gemini AI app on Mac, allowing users to interact with the AI assistant without switching windows. The app features a floating chat bubble that can be accessed with a shortcut, and can pull information from the user's current window to help answer questions. The app is similar to Apple's upgraded Spotlight feature.

Allbirds, a former shoe brand, has sold its assets for $39 million and is pivoting to become NewBird AI, a cloud solutions provider. The company has received a $50 million investment and plans to offer GPU-as-a-Service. This drastic pivot aims to capitalize on the growing AI sector.

Reid Hoffman, LinkedIn co-founder and venture capitalist, has expressed support for 'tokenmaxxing', a concept that tracks employee AI usage, stating it's a good idea to monitor token spend as a proxy for embracing AI tools. However, critics argue it's not a viable measure of productivity, akin to ranking people based on spending. Hoffman suggests pairing token tracking with understanding how tokens are used.

Former President Donald Trump has been posting AI-generated fan art depicting him as Jesus, sparking controversy and raising questions about the intersection of technology and politics. The images have been shared on Trump's social media platform, Truth Social. The move has drawn attention to the potential implications of AI-generated content on public discourse.

Grammy-nominated singer-songwriter Aloe Blacc is bootstrapping a cancer drug platform after being unable to fund research for a COVID solution despite being vaccinated and boosted. He found that philanthropy has limitations in advancing science through clinical trials. Blacc is now targeting pancreatic cancer with his new biotech venture.