NerdNews

Microsoft is integrating OpenClaw-like features into its Microsoft 365 Copilot tool, aiming to provide better security controls for enterprise customers. The new features will enable an agent to complete multistep tasks over long periods, with the goal of showcasing this technology at the Microsoft Build conference in June.

Google released Gemma 4, a family of models designed to support local, agentic AI for Android development. The models enable on-device AI inference, reducing reliance on cloud-based services and improving data privacy. Gemma 4 includes three models, with the most powerful one requiring 24GB of RAM and 17GB of storage, suitable for coding assistance in Android Studio. The models are up to 4x faster than previous versions and use up to 60% less battery.

Lyft implemented an AI-driven localization system, reducing translation turnaround from days to minutes. The system uses large language models and human review to ensure quality and cultural relevance.

Meta is building an AI version of Mark Zuckerberg to engage with employees. The AI character is being trained on the CEO's mannerisms, tone, and publicly available statements. This effort is part of Meta's broader push to develop 'personal superintelligence' and catch up with rivals in building cutting-edge AI models.

The 2026 AI Index Report from Stanford's HAI finds AI capability outpacing responsible AI, with safety benchmarks lagging and incidents rising sharply. AI models have become proficient in programming, but struggle with tasks like reading analog clocks and household chores. The report highlights concerns over AI's impact on elections and personal relationships, with 64% of the American public expecting AI to reduce job availability.

Mariia Bulycheva discusses using Graph Neural Networks (GNNs) to improve Zalando's recommender system, including data preparation, training process, and pitfalls. GNNs can model complex user interactions and item relationships, providing more accurate recommendations.

Spring Boot 4 introduces modularized auto-configurations, improving startup times. Spring Framework 7 adds built-in retry support and API versioning. The Spring team sees AI coding tools as transformational and plans to support them.

Anthropic has released Claude Mythos Preview, a capable frontier model with improvements in reasoning, coding, and cybersecurity. However, access is restricted to a consortium of tech companies through Project Glasswing due to its potential to autonomously discover and exploit zero-day vulnerabilities.

A leaked memo from OpenAI's chief new revenue officer reveals a new strategy to attack rival Anthropic, questioning its revenue figures and claiming it overstates its financial situation. OpenAI believes Anthropic's revenue is $8 billion less than reported, putting it behind OpenAI's $24 billion run rate.

Microsoft is testing OpenClaw-like AI bots for its Copilot AI assistant to make it an always-on agent that completes tasks on behalf of users. The company aims to show off some of these features during its Build conference and is exploring safer versions of the tool.

Kepler Communications has launched the largest orbital compute cluster, boasting 40 Nvidia Orin edge processors onboard 10 operational satellites. The company has partnered with Sophia Space to test its unique orbital computer, and plans to provide network services for other satellites and drones. This technology has the potential to revolutionize data processing in space and provide faster responsiveness for applications such as synthetic aperture radar.

Qualcomm Technologies has expanded its strategic partnerships with Bosch and Snap Inc to develop advanced driver assistance systems and power future generations of immersive eyewear. The collaboration with Bosch aims to address the auto industry's need for intelligent vehicle technology, while the partnership with Snap Inc focuses on developing advanced eyewear that integrates digital experiences into the physical world.

A critical vulnerability in the wolfSSL SSL/TLS library can weaken security via improper verification of the hash algorithm or its size when checking Elliptic Curve Digital Signature Algorithm (ECDSA) signatures, allowing attackers to force a target device or application to accept forged certificates for malicious servers or connections.

Leaked Steam client update files reference 'SteamGPT', suggesting Valve may use AI to streamline evaluations of in-game incidents and suspicious accounts, potentially utilizing generative pre-trained transformers for tasks like label generation and account summarization.

Cloudflare is rebuilding its Wrangler command-line tooling to support AI agents, adding new commands and features to improve usability and functionality. The company is releasing a technical preview of the updated Wrangler CLI, which includes a new TypeScript schema to define APIs, CLI commands, and context. Cloudflare is also introducing Local Explorer, a feature that allows users to inspect Cloudflare Workers and their bindings.

Operation Atlantic, a joint effort between UK, US, and Canadian law enforcement, has seized $12m in crypto losses and identified over 20,000 victims of 'approval phishing' scams. The operation also disrupted multiple fraud networks and identified $33m in additional crypto fraud.

AWS has launched a Sustainability console, providing a standalone service for carbon emissions reporting with its own permissions model, API, and Scope 1-3 emissions data. The console is available at no additional cost and includes historical data back to January 2022.

Pure Data Centres Group has partnered with AVK to create a self-powered datacentre in Dublin, using a microgrid to overcome grid constraints. The datacentre is powered by liquid natural gas and sustainable hydrotreated vegetable oil, with a goal of achieving net-zero operations by 2040. The project aims to provide a blueprint for future on-demand microgrids and reduce demand on the grid.

OpenAI is rotating potentially exposed macOS code-signing certificates after a GitHub Actions workflow executed a malicious Axios package. The company found no evidence of certificate compromise but is taking precautions. macOS users must update their apps to the latest versions by May 8, 2026.

Anthropic reduced the TTL for Claude Code prompt cache from 1 hour to 5 minutes, leading to faster quota depletion. Users report issues with cache rebuilding and misses, despite Anthropic's claims that the change should not increase costs.