Google DeepMind announced the general availability of the Interactions API, which serves as their primary interface for Gemini models and agents. Since its public beta launch in December 2025, this API has become a favored method for developers to build applications with Gemini.

With this GA release, the API now features a stable schema and introduces significant new capabilities such as Managed Agents, background execution, and the upcoming Gemini Omni. Developers can interact with models or run agents with just a few lines of code, and long-running tasks can be executed asynchronously by setting background=True.

Key updates include Managed Agents, which provide a remote Linux sandbox for agents to reason, execute code, browse the web, and manage files, along with tool improvements that now return images alongside text. The API also supports media generation, including image generation with Nano Banana 2, music with Lyria 3, and expressive speech with multi-speaker TTS. The Interactions API is now the default for Google AI Studio and the Gemini API, designed specifically for stateful, agentic workflows.

👉 Read the full article on Google AI Studio / blog.google

Meta and Google were found liable by a Los Angeles jury for the addictive design of their platforms, which caused harm to a young person's mental health, resulting in a significant compensation order. This ruling is groundbreaking as it focuses not on specific content, but on the architectural design elements like infinite scroll, autoplay, and recommendation systems that are engineered to keep users engaged.

The verdict shifts the debate from user-generated content to the responsibility of the companies that build and optimize these platforms, even when aware of potential risks. In Europe, the Digital Services Act (DSA) already mandates large platforms to assess and mitigate systemic risks, including those affecting mental health and minors, with features like TikTok's infinite scroll drawing regulatory scrutiny.

Moreover, the EU AI Act is also relevant, prohibiting certain AI uses that employ manipulative, deceptive, or subliminal techniques capable of significantly distorting a person's behavior and causing harm. Spain also has legal frameworks recognizing the potential for digital technologies to induce excessive behaviors and enhanced protections for minors, suggesting that a platform's design could be considered a "defective product" if it causes demonstrable health damage。

👉 Read the full article on Gizmodo en Español

OpenAI is reportedly launching a full-scale effort to patch open-source bugs as it aims to compete with Anthropic's "Mythos". This initiative suggests OpenAI's commitment to strengthening the robustness and reliability of its technological foundation amidst intensifying competition in the AI sector.

Identifying and fixing bugs in open-source software is crucial for ensuring the safety and efficiency of AI systems. OpenAI's focus on this area, especially as competitors like Anthropic develop advanced AI models, is likely a strategic move to maintain its technological edge.

👉 Read the full article on Wired