Edition: June 29, 2026

Meta hired contractors to pose as minors and test rival chatbots' responses to sensitive topics like suicide, sex, and drugs, sparking concerns over the company's methods. The project, known as Cannes, targeted chatbots from OpenAI, Google, and Character.AI. Meta defended the work as routine safety testing, but contractors questioned the approach.

Google's Gemini app now offers free personalized AI image generation to all eligible US users, allowing them to create unique images based on their interests and preferences. The feature, previously only available to paid subscribers, utilizes data from connected Google accounts to generate images. Gemini's user base has surpassed 750 million monthly active users, solidifying its position in the AI space.

Tidal has introduced a new policy on AI-generated music, labeling such tracks with an icon and removing their monetization starting July 15th, while prioritizing royalties for original human-created works. The platform will not pay royalties on music identified as 100% AI-generated. This move aims to protect artists and inform listeners.

California has partnered with Anthropic to provide state agencies with access to the AI chatbot Claude at half price, aiming to increase government efficiency. The deal includes training and support, and is part of Governor Newsom's effort to accelerate AI use in government. This move contrasts with the federal government's stance on Anthropic, which has been labeled a 'supply-chain risk'.

South Korean tech giants Samsung and SK Hynix have committed to investing over $550 billion to build new memory fabs and AI data centers, aiming to establish the country as a major AI power player. The investment is part of a national plan to boost the semiconductor and AI industries. The move is in response to the global shortage of memory chips, dubbed 'RAMageddon', driven by the growing demand for AI technology.

Researchers introduce DiScoFormer, a transformer model that can estimate both the density and score of a distribution from a set of data points in a single forward pass. This model leverages a shared backbone with two output heads and cross-attention to achieve state-of-the-art results. DiScoFormer has the potential to improve various applications, including diffusion-based generative models and Bayesian sampling.

A recent study found that managers who treat AI agents as coworkers perform worse and are less likely to take responsibility for the AI's output, highlighting the risks of anthropomorphizing AI tools. This trend is alarming as companies like Microsoft and Google release new AI management tools that frame agents as digital colleagues. The study suggests that AI agents should be optimized to improve human capabilities, not replace them.

Arena, a popular AI leaderboard provider, has reached $100 million in annualized run-rate revenue just eight months after launching its commercial service. The company's platform provides deep-dive performance analytics to model labs and enterprises, and its rapid growth shows the demand for its commercial offerings. Arena's success demonstrates the growing need for post-training optimization services in the AI industry.

TIDAL has introduced a new policy to prevent fully AI-generated music from being monetized on its platform, using automated tools to remove AI-generated music that impersonates artists. The policy aims to protect and reward organic creativity, with AI-generated tracks being tagged and ineligible for royalties or direct-to-fan sales. This move follows similar policies by other music streaming services like Spotify and Apple Music.

A Google expert explains that a full-stack AI approach integrates every layer of technology into one cohesive system, improving reliability and lowering costs. This strategy has been the foundation of Google's AI work, enabling the delivery of powerful and cost-efficient products. Google offers various platforms, including Google AI Studio and Gemini Enterprise Platform, to help developers build and automate AI systems.

US lawmakers propose a ban on selling Americans' health and location data to data brokers, including information shared with AI chatbots. The Health and Location Data Protection Act aims to protect sensitive information. Senator Elizabeth Warren and Representative Mary Gay Scanlon are set to introduce the revised bill.

Enterprise investment in AI is booming, with a focus on agentic AI to drive financial outcomes, and tech teams are increasingly putting agents to work to manage and coordinate workflows. However, agent readiness drops due to a lack of business context, highlighting the need for context-generation capabilities. As experience with agents deepens, experts expect agent confidence to accelerate.

Proception, a robot hand company founded by former Tesla engineer Jay Li, has settled a trade secret lawsuit with Tesla and announced an $11 million seed round to develop high-dexterity robotic hands. The company aims to become the top hand supplier to other robotics companies. Proception's innovative approach uses a sensor-laden glove to collect human hand interaction data, enabling more accurate and scalable robotic hand development.

NVIDIA researchers have developed ENPIRE, a software framework that enables self-improving robots to learn and adapt in the real world through autonomous experimentation and execution loops. The system has shown promising results in simple tasks such as dexterous manipulation, with robots achieving a 99% success rate. This technology has potential implications for the development of superintelligent robots.

HP has expanded its integration of OpenAI's Frontier platform to optimize enterprise workflows, resulting in significant operational gains in software engineering and cybersecurity remediation. The platform has enabled HP engineers to process large volumes of code requests and resolve software bugs more efficiently. This integration is expected to accelerate HP's enterprise operations and improve its partner channel ecosystem.

Advances in natural language processing are revolutionizing professional networking by enabling more personalized and relevant interactions. AI-powered tools are streamlining introductions, fostering contextual conversations, and helping users navigate digital spaces. This shift is transforming how users pursue and maintain professional connections, presenting both opportunities and challenges.

As DevSecOps teams release changes at a rapid pace, automated security testing has become crucial to catch routine flaws before they reach production. Tools like XBOW support this work by mapping application surfaces and testing likely attack routes. Automated testing helps security professionals provide better proof and faster handoffs to engineering teams.

xFusion has introduced a scalable enterprise AI computing model that transitions hardware from edge devices to data centres, offering a four-tier hardware portfolio to meet the needs of enterprise technology buyers. The portfolio includes personal edge processing devices, workgroup data containment appliances, and larger systems for corporate offices and facilities. This solution aims to address the challenges of practical production frameworks and data security in AI computing.

A Swiss startup, Flexion Robotics, has developed a system to train humanoid robots to perform complex office tasks autonomously, using a combination of AI systems and reinforcement learning. The robots can learn to open doors, climb stairs, and carry boxes, and even retrieve parcels and unpack them. This technology has the potential to revolutionize the humanoid market, estimated to be worth $150 billion by 2036.

Scam.ai has announced a partnership with Qualcomm and launched its Halo deepfake detection model, which can identify synthetic video in real-time during live video calls. The model operates on-device, ensuring user privacy and security. This technology is particularly useful for high-stakes video calls, such as those conducted by HR and recruiting teams or high-value executives.

Omen AI has developed a tiny spectrometer to monitor fluid health in real-time, helping data centers prevent bacterial outbreaks and costly downtime. The company has raised $31 million in a Series A round to further develop its solution. Omen AI's technology has the potential to optimize data center operations and reduce maintenance costs.

OpenAI is set to release a new hardware device related to its Codex coding tool on July 15th, featuring a square-shaped design with buttons, in partnership with Work Louder. The device appears to be a custom keyboard or macro pad designed to enhance the Codex experience. Further details are expected upon the official launch.

Cursor has launched a mobile app for guiding coding agents on-the-go, allowing users to interact with and spin up new agents from their phone. This move follows similar apps from Anthropic and OpenAI, reflecting a shift towards AI-based coding tools that abstract away from written code. The app ties into Cursor's 2.0 changes, which focus on independent coding agents.

The Download newsletter discusses the limitations of metrics in self-quantification and highlights AI-powered warning systems in India to prevent human-elephant clashes. Additionally, the US has allowed Anthropic to release its Mythos 5 model to trusted organizations.

Wimbledon is introducing new AI-powered features to its digital platforms, including an upgraded Match Chat assistant and a Key Moments feature, to enhance live match coverage. These features will be available through the Wimbledon app and website, and are part of a five-year digital transformation project with IBM. The updates aim to provide a more engaging and informative experience for fans.