Updated just now

Share this article

Related News

Google DeepMind Challenges Anthropic with New AI Coding Strike Team

Apr 21, 2026

Google DeepMind Challenges Anthropic with New AI Coding Strike Team

Google DeepMind has set up a 'strike team' to enhance its AI coding models and catch up with Anthropic's Claude tools. With leaders like Sergey Brin pushing this innovation, DeepMind aims to boost Gemini's capabilities to improve itself and dominate AI development.

Google DeepMindAI codingAnthropic

Attack on OpenAI CEO’s Home Highlights Rising Security Threats to AI Leaders

Apr 14, 2026

Attack on OpenAI CEO’s Home Highlights Rising Security Threats to AI Leaders

A concerning incident at the home of OpenAI CEO Sam Altman has alarmed the AI community, revealing a troubling development in security threats against AI leaders. An accused attacker not only targeted Altman but was found with a list of other AI executives, indicating a broader threat related to the contentious debates surrounding AI development and ethics.

OpenAISam AltmanAI security

Anthropic Steps Back: Claude Mythos AI Model Deemed 'Too Risky' for Public Eyes!

Apr 10, 2026

Anthropic Steps Back: Claude Mythos AI Model Deemed 'Too Risky' for Public Eyes!

In a daring move, Anthropic, a leading AI trailblazer, has decided not to roll out its latest AI marvel, Claude Mythos, to the eager public. Concerns over the model's potential high-stakes risks have prompted Anthropic to hit pause, reflecting on the possible safety threats in a rapidly advancing AI landscape. The decision underscores Anthropic's cautious approach towards deploying hyper-advanced tech into the wrong hands. Keep reading to discover what this means for the AI industry and global security!

AnthropicClaude MythosAI safety

Trusted testers get first Genie 4 access; Ultra and API rollout is still to come — Source: OpenTools Team / AI-assisted conceptual illustrationSource image

Trusted testers get first Genie 4 access; Ultra and API rollout is still to come

By OpenTools Team

Google’s Genie 4 preview lets testers steer scenes with prompts and adds generated audio. Wider Ultra and API access is still planned, with drift and consistency limits unresolved.

Trusted testers come before the wider rollout

Google announced Genie 4 on October 27 as a world‑model research preview for creating and steering interactive scenes. The company says it has begun rolling the model out to a limited group of trusted testers. Google AI Ultra access through Project Genie is still described as coming soon, alongside early‑access partners in the Gemini API and Google AI Studio.

That separates the model announcement from a broad consumer release. An Ultra subscription is not, on the reviewed announcements alone, proof that a person can already open Genie 4. The DeepMind post offers a waitlist, but gives no exact date for the wider rollout. It also does not establish a generally available API or an entitlement for every AI Studio user.

Prompts steer an evolving scene

Google describes controls for changing a simulation while it runs: moving the camera, changing the weather, or introducing a person or object with natural‑language prompts. The DeepMind demonstrations are described as recordings of real‑time interactions using keyboard navigation and camera controls alongside those prompts.

The creative distinction is between viewing a fixed clip and making choices inside a scene as it unfolds. Google says testers are exploring uses such as scene pre‑visualization and interactive prototypes. Those are reported uses of an experimental system, not evidence that it replaces a game engine, produces production‑ready footage or reliably reproduces a physical process. OpenTools has not used the preview or recreated the demonstrations.

Audio is generated with the world

The announcement says Genie 4 generates audio as part of the scene, including ambient sound, music and interactive dialogue. DeepMind describes characters responding to inputs or events in the world. These are the developer’s descriptions of the model’s intended capabilities; the supplied recordings are demonstrations rather than an independent evaluation of every possible prompt.

For someone planning a creative workflow, generated sound is also something to review. The announcement does not establish that dialogue, timing or environmental sound will remain accurate and consistent through an arbitrary sequence of edits. It does not provide the wider rollout’s final access terms, pricing or usage allowances in the material reviewed here.

Google identifies drift and consistency limits

DeepMind’s limitations section says visual details and object permanence can shift or drift as a session runs longer, after complex prompt edits, or during navigation. It also lists physical interactions, audio accuracy, legible text, spatial accuracy and long‑term consistency as areas of continuing research.

Those limitations belong beside the creative promise. A scene that looks convincing in a supplied example is not proof that a particular object, label or event will remain consistent through a longer session. The current announcement supports an experimental preview with steerable scenes and generated audio. It does not establish dependable simulation physics, a validated safety‑training environment or finished creative output without review.

Conceptual illustration created for OpenTools: a handmade paper landscape, movable cloud, navigation controller and speaker represent steering an imagined scene with sound. This is original illustrative artwork, not output from Genie 4, a Google DeepMind demonstration, a game interface or evidence of simulation quality or physics.

Sources

1.DeepMind post (deepmind.google)
2.announcement (blog.google)
3.imagedelivery.net (imagedelivery.net)

Tags

Google DeepMindGenie 4World models Creative AIAI previews