October 19, 2026
OpenAI appears to be testing a new image generation model internally codenamed "Aurora," with three alpha variants briefly appearing in the company's API before being pulled. The models have since shown up in Arena-style blind comparison tests, suggesting OpenAI is gathering quality benchmarks ahead of a wider release.
New research finds that AI-edited photos and videos can cause people to form false memories of events - and that exposure to manipulated imagery tends to make viewers more confident in those incorrect recollections. The findings raise fresh concerns about the psychological impact of AI-generated and AI-altered visual media as it becomes increasingly difficult to distinguish from authentic footage.
Kling AI has updated its video generator to version 4.0, bringing stereo audio, multilingual lip sync, clips up to 30 seconds, and support for up to ten keyframe images. A lighter Flash variant accompanies the release for users who prioritize speed over maximum quality. Benchmark comparisons suggest Kling 4.0 is competitive with peers like Wan 3.0 and Seedance 2.5, though it does not clearly outpace them.
Sound stages built during the streaming boom are sitting vacant across the US and UK, and private equity firms are increasingly converting them into data centers rather than waiting for Hollywood production to recover. The shift reflects a broader reallocation of capital away from film infrastructure and toward AI computing - driven by post-strike production slowdowns, tighter streaming budgets, and the rising demand for server space. Several high-profile studio projects have already been halted
Google has launched a public-facing website for SynthID, its AI watermarking and detection tool, allowing anyone to check whether an image has been created or significantly altered by AI. The tool works by reading watermarks embedded by Google's own image generation systems, which limits its coverage but makes it a practical first check for content from those pipelines.
