Latest Weekly News

Introducing the OpenAI Safety Bug Bounty program

OpenAI has launched a Safety Bug Bounty program focused on uncovering AI abuse patterns and practical model safety risks.

Read original source ↗

Mistral releases a new open source model for speech generation

Mistral has launched an open-source speech generation model aimed at enterprise voice agents and customer engagement use cases.

Read original source ↗

Protecting people from harmful manipulation

Google DeepMind outlines harmful manipulation risks in AI and introduces new safety measures across sensitive areas like health and finance.

Read original source ↗

How Autonomous AI Agents Become Secure by Design With NVIDIA OpenShell

NVIDIA presents OpenShell as a security-by-design approach for autonomous AI agents that can act across tools, code, and enterprise workflows.

Read original source ↗

VCs are betting billions on AI’s next wave, so why is OpenAI killing Sora?

AI investment is accelerating, but resistance to data center expansion shows the next phase of AI will be shaped by real-world limits.

Read original source ↗

OpenAI shuts down Sora while Meta gets shut out in court

AI’s expansion is meeting real-world resistance, from land use disputes to courtroom setbacks for major tech players.

Read original source ↗

Google launches Lyria 3 Pro music generation model

Google has introduced Lyria 3 Pro, a new music generation model focused on longer tracks and more control across its AI products.

Read original source ↗

Powering product discovery in ChatGPT

ChatGPT is adding richer shopping and product comparison features, with merchant integration designed to improve discovery inside the chat experience.

Read original source ↗

OpenAI abandons yet another side quest: ChatGPT’s erotic mode

OpenAI appears to be dropping ChatGPT’s erotic mode, the latest sign it is trimming side projects and narrowing focus.

Read original source ↗

You can now transfer your chats and personal information from other chatbots directly into Gemini

Google is adding tools to move chats and personal data from other chatbots into Gemini, making switching easier.

Read original source ↗

Gemini 3.1 Flash Live: Making audio AI more natural and reliable

Google DeepMind says Gemini 3.1 Flash Live improves voice AI with lower latency and better precision for more natural audio interactions.

Read original source ↗

Conntour raises $7M from General Catalyst, YC to build an AI search engine for security video systems

Conntour raised $7M to build AI search for security video, letting teams query camera feeds in natural language.

Read original source ↗

Inside our approach to the Model Spec

OpenAI outlines how its Model Spec guides AI behavior, aiming to balance safety, user autonomy, and accountability.

Read original source ↗

Helping developers build safer AI experiences for teens

OpenAI has introduced teen safety policies for developers using gpt-oss-safeguard to better manage age-specific risks in AI applications.

Read original source ↗

A New Framework for Evaluating Voice Agents (EVA)

Hugging Face highlights EVA, a new framework designed to evaluate voice agents in a more structured and practical way.

Read original source ↗

Lyria 3 Pro: Create longer tracks in more

Google DeepMind introduced Lyria 3 Pro, focused on longer music tracks with stronger structural awareness and broader product integration.

Read original source ↗

Update on the OpenAI Foundation

OpenAI says its foundation will invest at least $1 billion in health, economic opportunity, AI resilience, and community programs.

Read original source ↗

Advancing Open Source AI, NVIDIA Donates Dynamic Resource Allocation Driver for GPUs to Kubernetes Community

NVIDIA is donating a dynamic GPU resource allocation driver to Kubernetes, aiming to improve how AI infrastructure is managed in open source environments.

Read original source ↗

Data centers get ready — the Senate wants to see your power bills

Two senators want closer reporting on how data centers consume electricity and what that means for grid reliability and energy planning.

Read original source ↗

Into the Omniverse: NVIDIA GTC Showcases Virtual Worlds Powering the Physical AI Era

NVIDIA GTC highlighted how Omniverse, OpenUSD and virtual worlds are becoming core tools for scaling physical AI across robots, vehicles and factories.

Read original source ↗

A ‘pound of flesh’ from data centers: one senator’s answer to AI job losses

As AI job-loss fears rise, Sen. Mark Warner proposes taxing data centers to help workers manage the transition.

Read original source ↗

Mercor competitor Deccan AI raises $25M, sources experts from India

Deccan AI raised $25M by focusing on India-based experts, betting that tighter workforce control can improve quality in AI training.

Read original source ↗

The AI skills gap is here, says AI company, and power users are pulling ahead

Anthropic says AI is not replacing jobs yet, but skilled users are already gaining an advantage that could widen workforce inequality.

Read original source ↗

The Future of AI Is Open and Proprietary

NVIDIA argues AI’s future will rely on a mixed ecosystem of open and proprietary models, from broad platforms to specialized tools.

Read original source ↗

Bernie Sanders and AOC propose a ban on data center construction

Sanders and Ocasio-Cortez want to pause new data center construction until Congress sets broader AI rules.

Read original source ↗

Anthropic wins injunction against Trump administration over Defense Department saga

A federal judge ordered the Trump administration to reverse recent restrictions imposed on Anthropic in a Defense Department dispute.

Read original source ↗

Google unveils TurboQuant, a new AI memory compression algorithm — and yes, the internet is calling it ‘Pied Piper’

Google’s new TurboQuant algorithm could cut AI working memory by up to 6x, but for now it remains an early-stage research project.

Read original source ↗

Creating with Sora Safely

OpenAI says Sora 2 and its creation app were designed with safety built in, aiming to address both model and platform risks from the start.

Read original source ↗