Stay ahead on AI.
All in ONE place.

Thinking Machines has released Inkling, an open-weights Mixture-of-Experts model with 975B total parameters and 41B active, pretrained on 45 trillion tokens of text, images, audio, and video. It supports a 1M-token context window and features controllable thinking effort, allowing developers to trade performance for token efficiency. Inkling is available for fine-tuning on Tinker, the company's customization platform. A smaller variant, Inkling-Small, with 12B active parameters, is also previewed alongside the main release.

A hacker who breached Suno AI shared source code with 404 Media revealing the music generation company scraped millions of songs from YouTube Music, Deezer, Genius, Pond5, Jamendo, and Freesound, totaling over 380,000 hours of audio. The code shows Suno used proxy service Bright Data to scrape YouTube and also targeted roughly 1 million hours of podcasts. The hacker also accessed hundreds of thousands of customer records including Stripe payment details. Suno confirmed a 2025 security incident but said no sensitive personal information was compromised.

OpenAI has launched Codex Micro, a $230 hardware device made in collaboration with keyboard maker Work Louder, available in limited quantities via Supply Co. The square pad features 13 mechanical switches, a joystick, dial, and touch sensor, and is designed to work with OpenAI's Codex coding platform. Six frosted keys display live status of Codex agent threads using color indicators, while command keys handle actions like push-to-talk and accepting changes. All controls are configurable through the ChatGPT desktop app.

OpenAI's Chief Global Affairs Officer Chris Lehane outlined a "reverse federalism" strategy for AI safety, where states like California, New York, and Illinois pass aligned frontier AI legislation to build a de facto national standard. The framework centers on three elements: documented safety assessments with public disclosure, incident reporting, and independent audits. OpenAI also supports a federal testing framework the Trump Administration aims to finalize by early August, and backs a US-led international forum for global AI governance standards.

OpenAI has released GPT-Red, an automated red-teaming model trained via self-play reinforcement learning to find and exploit prompt injection vulnerabilities. GPT-Red is used to adversarially train production models, resulting in GPT-5.6 Sol achieving 6x fewer failures on OpenAI's hardest direct prompt injection benchmark compared to models from four months prior. In testing, GPT-Red outperformed human red-teamers 84% to 13% on novel scenarios and successfully manipulated a live office vending machine agent to alter prices and cancel orders.

George Lucas has voiced support for AI in filmmaking, telling A Rabbit's Foot that artificial intelligence makes movies "much easier" to produce and comparing resistance to it to opposing the automobile. "There's nothing you can do about it. That's progress, it's the future," Lucas said. The Star Wars creator, who sold Lucasfilm to Disney for $4 billion in 2012, also criticized Hollywood's over-reliance on focus groups, arguing studios let audiences make movies instead of passionate filmmakers.

AI safety researcher Cereblab found that Grok Build, xAI's command-line interface, was uploading entire user repositories including full Git histories to a Google Cloud Storage bucket, even when instructed not to open files. One user had their entire home directory, including SSH keys and password manager databases, uploaded. xAI fixed the issue via a server-side flag, disable_codebase_upload set to true. Elon Musk promised all previously uploaded data will be deleted, though Cereblab notes the /privacy command was not what actually stopped the uploads.

OpenAI's first consumer device will be a screenless, movable smart speaker designed as a humanlike AI companion. The battery-powered device can be carried room to room, includes a camera and sensors, and will use GPT-Live voice mode to control smart-home appliances, answer questions, and learn its owner's habits over time. Built with help from Jony Ive's LoveFrom studio and former Apple designer Evans Hankey, the speaker is planned for unveiling in 2026 with a 2027 release, though Apple's trade-secrets lawsuit could delay sales.

OpenAI has responded to a trade secrets theft lawsuit filed by Apple, denying the complaint has merit. In an official statement, OpenAI said: 'While we take these allegations seriously, we're not aware of any evidence that this complaint has merit.' The company added that it believes in fair competition and employee freedom of movement, stating it remains focused on building innovative technology that empowers people everywhere. No further details about the lawsuit's specific allegations were disclosed.

Google is celebrating the 25th anniversary of Google Images by introducing two new features. A redesigned, personalized gallery offers a dynamic, real-time browseable homepage tailored to user interests, rolling out on desktop in the U.S. Additionally, image generation is coming to AI Overviews in Search, powered by Google's Nano Banana model, letting users create custom visuals from text prompts. Both features will roll out over coming weeks in English across regions that support AI Mode.