AI Dubbing on Instagram, GPT-5.2, Disney vs. Google, and AI Hacking Bots
This Week in AI Newsletter: 12/12/2025
OpenAI released GPT-5.2 (Instant, Thinking, and Pro), their new flagship model on the 10 year anniversary of ChatGPT. The model has improved abilities to handle spreadsheets, create slideshows, coding and many other features. What is obvious is that GPT-5.2 did not blow any other models out of the model, they did outperform some other in some benchmarks, but 2 tests online make it clear that OpenAI is still behind Google: One was design, and another was visual understanding. More here.
ElevenLabs partnered with Meta to bring voice to Instagram, Horizon and more starting with AI dubbing. Some early use cases are dubbing Reels in local languages, generating music and character voices in Horizon. They will be partnering with 3 models: Text to Speech, Dubbing, and Music. More here.
Design and coding are merging into a one process, and Cursor now lets you design directly inside your codebase. You can visually adjust UI elements in place while it writes the corresponding code automatically. By moving design changes straight into production code, Cursor weakens Figma’s role as a separate design layer that requires handoff and translation into code. More here.
Google introduced Disco, a Gemini-powered experiment that turns your browser tabs into custom mini web apps called GenTabs. It can suggest apps for things like studying, meal planning, or trip planning, and it is starting as a limited Google Labs test with a macOS waitlist. More here.
Google unveiled a revamped Gemini Deep Research agent built on Gemini 3 Pro, and for the first time is letting developers embed its “deep research” capabilities into their own apps via a new Interactions API. The launch looks like a well-timed shot across the bow at OpenAI, landing the same day OpenAI released GPT-5.2. More here.
1X struck a strategic partnership with Swedish investor EQT to make up to 10,000 of its Neo humanoid robots available to EQT portfolio companies, targeting factories, warehouses, logistics, and other industrial uses between 2026 and 2030. The deal shifts away from Neo’s consumer “home robot” positioning and suggests industrial deployments may be the faster path to scale for humanoid robots. More here.
Disney sent Google a cease-and-desist letter alleging “massive” copyright infringement, claiming Google’s AI models and services commercially distribute unauthorized images and videos of Disney characters (including titles like Frozen, The Lion King, and Moana). Google said it will “engage” with Disney and pointed to its use of public web data plus copyright controls like Google-extended and YouTube’s Content ID. This comes the same day that Disney partnered with OpenAI. More here.
Runway AI launched its first “world model,” GWM-1, which predicts video frame by frame to simulate how the world behaves. It also updated its Gen 4.5 video model to generate videos with built-in audio, including dialogue and background sound. More here.
Stanford researchers tested an AI hacking bot called Artemis on the university’s engineering network and found it beat nine of 10 professional penetration testers at finding vulnerabilities while costing under $60 per hour compared with roughly $2,000 to $2,500 per day for humans. The system was not perfect, producing about 18% false positives and missing an obvious bug that most human testers caught. More here.