The last month has been a milestone month for artificial intelligence, with breakthroughs, launches, and regulations shaping the global AI landscape. From DeepMind’s Genie 3 world model, seen as a step toward AGI, to ElevenLabs’ commercial-ready AI music generator, innovation pushed forward at speed.
OpenAI joined AWS for the first time, Musk’s xAI launched Grok Imagine for AI-powered images and video, and the EU AI Act began applying to high-risk general-purpose models.
Here’s a closer look at the biggest AI developments you need to know.
Genie: DeepMind’s Bold Move Toward Artificial General Intelligence
Google DeepMind’s Genie, spanning from its earliest iteration through the breakthrough Genie 3, marks a bold leap in the development of AI world models, bringing the dream of AGI closer to reality. Introduced in early 2024, the original Genie model was envisioned as a generative interactive environment trained in an entirely unsupervised manner using unlabeled internet videos. It could generate an endless variety of action-controllable virtual worlds, and let users act within them on a frame-by-frame basis, despite having no ground-truth action labels.
This capability unlocked the possibility of training generalist agents capable of mimicking behaviors from unseen scenarios, which is fundamental for AGI development. Building upon that foundation, Genie 3 was unveiled last month, and takes matters to a new level.
It can generate fully interactive 3D worlds in real time at 720p resolution, and maintain consistency and memory for several minutes, a significant leap over prior versions that lasted mere seconds.
Key Capabilities That Define Genie 3
- Promptable World Events
Users (or AI agents) can dynamically modify the environment—altering weather, adding objects, or introducing new characters—via simple text prompts, enabling rich “what-if” scenarios.
- Visual Memory & Coherence
Genie 3 preserves the world’s state, even when elements are outside the frame, allowing long-term immersion and consistent interaction.
- Real-World Physics Learning
Instead of hardcoded physics engines, Genie 3 learns physical laws through autoregressive modeling, providing more organic, emergent behaviors.
- Agent Training in Virtual Worlds
When paired with DeepMind’s generalist agent SIMA, Genie 3 enables complex task execution (e.g. directing an agent to “approach the bright green trash compactor” in a simulated warehouse) through immersive, consistent environments.
- Broad Domain Flexibility
Genie 3 generates both realistic and fantastical worlds, from warehouses and ski slopes to historical recreations, making it a versatile tool across research, education, and entertainment.
Bridging to AGI
DeepMind frames Genie 3 as a critical stepping stone toward AGI. By enabling embodied agents to explore and learn from richly simulated environments, AI systems can develop planning, exploration, and reasoning skills reminiscent of human learning, not just reacting but acting with foresight.
Demis Hassabis sees this as a return to DeepMind’s agent-based roots to “one AI playing in the mind of another AI” inside Genie-generated worlds, a compelling vision of layered intelligence fueling AGI development.
Google DeepMind’s Genie, culminating in the advanced Genie 3 model, embodies a transformative step toward AGI by enabling AI systems to operate within richly interactive, memory-rich world models, paving the way for embodied, general-purpose intelligence.
From Prompts to Playlists: ElevenLabs’ New AI Music Tool
AI audio startup ElevenLabs has launched an AI music generator that it says is cleared for commercial use, marking its biggest expansion since its founding three years ago. Known for its industry-leading AI voice tools, including text-to-speech, conversational bots, and translation systems, ElevenLabs is now moving into generative music.
To showcase the model’s potential, the company released samples of AI-generated songs. One track featured a synthetic rapper describing a journey “from Compton to the Cosmos,” echoing the lyrical style of hip-hop icons like Dr. Dre and Kendrick Lamar. While impressive, such output also raises questions about cultural appropriation and authenticity in AI-generated art.
The launch comes amid heightened scrutiny of AI music platforms. In 2024, Suno and Udio were sued by the Recording Industry Association of America (RIAA) for allegedly training on copyrighted music without authorization. Those lawsuits highlight the risks startups face when entering the music space without clear rights frameworks.
To avoid similar pitfalls, ElevenLabs announced licensing deals with Merlin Network and Kobalt Music Group, which represent artists such as Adele, Nirvana, Bon Iver, and Childish Gambino. Crucially, these agreements are opt-in, meaning artists must give explicit permission for their work to be used. According to Kobalt, the deals include revenue sharing, safeguards against misuse, and fair terms for musicians.
By aligning with rights holders from the outset, ElevenLabs is positioning itself as a more ethical alternative in AI music. The company’s claim that its tool is commercial-ready could make it attractive for creators, advertisers, and developers seeking reliable AI-generated soundtracks and songs.
ElevenLabs’ expansion underscores both the creative possibilities and the cultural challenges of AI music, setting the stage for broader debates about technology’s role in reshaping artistry.
OpenAI Models Arrive on AWS in Competitive Shake-Up
For the first time, OpenAI models are available on Amazon Web Services (AWS), marking a major shift in the cloud AI landscape. OpenAI unveiled two open-weight reasoning models, comparable to its proprietary o-series, while AWS confirmed the models will be accessible through Amazon Bedrock and SageMaker.
The models, also downloadable via Hugging Face, are being offered with OpenAI’s explicit approval. AWS described the move as similar to its earlier distribution of DeepSeek-R1, but the inclusion of OpenAI is a far more symbolic win.
Until now, AWS’s AI roster has leaned on models from Anthropic (Claude), Cohere, and Mistral. Meanwhile, Microsoft Azure has dominated as OpenAI’s most important cloud partner, drawing scrutiny from investors who questioned whether AWS was falling behind in generative AI. During Amazon’s recent earnings call, CEO Andy Jassy faced tough questions about slower AI momentum compared to Microsoft and Google.
The partnership gives AWS customers a new way to build and experiment with OpenAI-powered apps, while strengthening OpenAI’s bargaining power as it renegotiates its deal with Microsoft. It also positions OpenAI against Meta, which has scaled back plans to fully open-source future models, while these new releases come under the permissive Apache 2.0 license.
The move follows reports that Oracle secured a $30 billion annual deal to provide OpenAI with data center services, underscoring the company’s strategy to diversify beyond Microsoft. For AWS, finally hosting OpenAI models represents both a strategic win and a timely rebuttal to Wall Street’s skepticism.
xAI Launches Grok Imagine for AI Images and Videos
xAI has launched Grok Imagine, a new tool in its Grok app that lets users create AI-generated images and videos. The feature debuted over the weekend on iOS for Heavy and Premium Plus subscribers and is now available to Heavy users on Android.
Grok Imagine lets users create images from text prompts and transform those images into short, sound-backed video clips. Users can upload photos or generate visuals directly, then select modes like Custom, Spicy, Fun, or Normal to create videos. Early testers, including Mashable, found the tool easy to use but noted that results were “fine” rather than exceptional.
Excitement around the launch has been amplified on X (formerly Twitter), where users are sharing their creations and xAI founder Elon Musk is actively retweeting them. Musk praised the tool as a “meme motherlode” and hinted at its NSFW capabilities, adding that Grok Imagine should improve “almost every day.”
The release follows Grok 4, xAI’s latest model, and places the company in direct competition with other AI video and image tools such as Midjourney’s video generator, and Google’s Veo 3. However, Grok Imagine appears to lack the same safeguards against deepfakes and misuse that rivals have implemented, raising questions about responsible deployment.
With Grok Imagine, xAI is betting on viral, user-driven content to boost Grok’s appeal and expand its footprint in the crowded generative AI space.
The EU AI Act: The World’s First Comprehensive AI Law
The European Union’s Artificial Intelligence Act (EU AI Act), described as the world’s first comprehensive AI law, is gradually coming into force across the EU’s 27 member states. Covering over 450 million citizens, it sets rules for both European and foreign companies developing or deploying AI systems.
Why the EU AI Act Exists
The Act establishes a uniform legal framework to avoid fragmented national rules, ensuring the free movement of AI-based goods and services across the EU.
Purpose and Guiding Principles
Its goal is to promote human-centric, trustworthy AI while safeguarding democracy, public safety, fundamental rights, and environmental protection. Lawmakers aim to balance innovation with harm prevention.
The Act introduces a tiered system:
- Unacceptable risk applications, like mass facial recognition scraping, are banned.
- High-risk uses, such as biometric identification or credit scoring, face strict oversight.
- Limited-risk scenarios carry lighter transparency obligations.
Rollout and Deadlines
Since August 2, 2025, rules now apply to general-purpose AI (GPAI) models with systemic risk, affecting players like Google, Meta, Anthropic, and OpenAI. Existing models have until August 2027 to comply. Non-compliance can lead to fines of up to €35 million or 7% of global turnover for prohibited uses, and €15 million or 3% for GPAI violations.
Industry Response and Global Impact
Reactions to the EU AI Act within the tech sector have been sharply divided. Supporters view it as a necessary framework to build trust, ensure safety, and create a level playing field for innovation. However, some companies, including Meta and Mistral AI, argue the rules risk slowing Europe’s AI progress and adding legal uncertainty.
Globally, the Act sets a powerful precedent, influencing how other regions may regulate AI. Its extraterritorial scope means international players like OpenAI, Google, and Anthropic must also adapt, reinforcing Europe’s role as a standard-setter.
