NerdNews

Researchers at White Circle found that new AI models from Google and OpenAI can be easily tricked into generating explicit content, including nudes. The researchers used a novel attack method called 'Red-teaming via Iterative Strategy Evolution' (RISE) to bypass the safety filters of the models. The findings suggest that the defenses of these models are weaker than acknowledged, and that mandatory independent safety testing is needed to prevent the misuse of AI technology.

Disney has offered $20 million to acquire some of the technology behind Character.AI, a chatbot maker that has been linked to controversy, including a lawsuit alleging its chatbots encouraged a 14-year-old boy to commit suicide. The move comes after Disney previously criticized Character.AI for stealing content and harming kids. Character.AI has filed for bankruptcy and is looking to sell its assets, with Disney potentially taking over the platform.

Google's Gemini 4 AI agents have successfully beaten Minecraft's final 'Ender Dragon' foe without human intervention. The four agents worked together, sharing tips and threats, and recovered items from dead teammates. They achieved this feat in 27.23 hours, using 1.35 billion input tokens and incurring a cost of $810 at Gemini's list price.

Nscale and Anthropic are reportedly in talks for a $45 billion compute deal, their second in a month. The deal would be a significant investment in AI compute infrastructure, following a string of large deals by Anthropic. The companies are also pursuing IPOs, with Anthropic expected to go public this year and Nscale seeking to go public despite investor pushback over its valuation.

Amazon Web Services has released TorchNeuron, a tool that allows AI startups to easily move their AI apps onto Amazon's AI chip, Trainium. This move challenges Nvidia's dominance in the AI hardware market. TorchNeuron works with PyTorch, a popular open-source toolkit, and lets AI apps run on Trainium with just one line of configuration change. Amazon argues that this will give AI startups access to its entire cloud and save them money.

Agentic AI introduces a new paradigm in distributed systems, focusing on decision decomposition rather than service decomposition. This shift enables more adaptive and context-aware systems but also introduces new complexities and failure modes. Key takeaways include the need for bounded autonomy, evaluation loops, and hybrid deterministic fallback systems to ensure reliability and observability in agentic architectures.

Buildkite has introduced pipeline-scoped secrets, new Model Context Protocol tools, and expanded OpenTelemetry data. The new secrets view restricts access to secrets to the current pipeline, and the MCP tools validate pipelines, diagnose failed builds, and repair webhooks. Buildkite has also added Linux arm64 support to its GitHub Actions compatibility layer.

.NET 11 Release Candidate 2 is now available with a go-live license, bringing updates across libraries, runtime, F#, ASP.NET Core, .NET MAUI, and Windows Forms. This release includes new features, improvements, and bug fixes, and is supported in Visual Studio 2026 Insiders and Visual Studio Code with the C# Dev Kit.

NestJS 12 introduces ESM packages, standard schema validation, and native observability. It requires Node.js 20.19+ or 22.12+, and the CLI now generates ESM projects by default. Validation is improved with support for Standard Schema compatible objects, and observability is enhanced with a new SDK. The upgrade process involves installing the latest CLI and running a migration command.

Epic Games has released a new 'Preview' build of its Epic Games Launcher, which was built using 'vibe coding' with AI. The new launcher is faster and more responsive than the current one, with a new look and features such as a Recent Games list and a centralized downloads hub.

Microsoft introduced new security features to protect enterprises from AI agent threats, warning that 'frontier models can't be trusted' due to their unpredictability and potential access to sensitive data. The company's solution includes Guardian Agents, a feature to monitor AI agent behavior, and an agent detection feature in Microsoft Defender.

Tenstorrent, a chip startup focused on inference, has raised $1.5 billion at a $6.2 billion valuation to scale open-weight AI chips, competing with Nvidia's dominance in the AI hardware market.

Town, an AI assistant startup, has reached a unicorn valuation after just 4 months of public launch, with a $108 million Series B funding round. The company's AI assistants, called Townies, help employees with tasks by connecting to tools like emails and calendars.

Rivian's CEO, RJ Scaringe, believes autonomy will allow for dramatically different vehicle designs. With autonomy, Rivian plans to build vehicles that can operate without human intervention, enabling new form factors and experiences. Scaringe envisions a future where personally owned autonomous vehicles will still be the norm, with 80-90% of miles driven in personally owned vehicles.

Laplace Intelligence is a startup that uses automated red-teaming to test robots and self-driving cars before they hit the real world. The company's simulation environment is paired with an AI agent that constantly probes for weaknesses and failure points, ensuring that physical AI is safe in simulation before deployment.

A security breach at OpenCode, an open-source AI agent framework provider, has exposed API keys linked to users of Anthropic, OpenAI, and Google AI models. The breach occurred on October 16 and affected the Console service, which allows teams to share access to AI models. OpenCode has notified affected users and recommended steps to mitigate the issue, including checking for suspicious activity, creating replacement keys, and revoking potentially exposed keys.

Indie game developers are becoming increasingly cautious about sharing their work due to the rise of AI-powered game clones. These clones can be created quickly and efficiently, making it difficult for original developers to protect their ideas. Some developers are delaying public demos or sharing their work only in closed sessions to avoid being cloned.

The Federal Trade Commission (FTC) is probing OpenAI and Anthropic over potential consumer harms caused by security incidents involving their AI models. The FTC sent civil investigative demands to the companies, seeking information on recent incidents, internal security measures, and marketing materials. The investigation may lead to fines or litigation if the FTC finds evidence of wrongdoing.

Amazon has signed an agreement with X-energy and American Electric Power to fund a new nuclear energy facility in Indiana, providing 960 megawatts of carbon-free energy for the grid and Amazon's data centers.

Vibe coding, powered by AI, is enabling hobbyist hackers to create custom firmware for music gear, breathing new life into old instruments and adding new features to existing ones. This has led to an explosion of custom firmware and hacks for synths, samplers, and grooveboxes, with many developers using AI tools like Claude to speed up the development process.