Episode Details

A plaintiff faces sanctions for hiding prompt injections in court filings, Google allows watermark removal, Apple partners with Alibaba, OpenAI adds an ultrafast tier, and Alibaba releases Qwen 3.8.

1. Plaintiff sanctioned for hiding invisible prompt injections in court filings — A Connecticut plaintiff lost electronic filing privileges after embedding invisible instructions into legal documents. The court established a legal precedent against attempting to manipulate automated review systems.
2. Google lets users remove visible watermarks from Gemini AI generations — A new toggle allows creators to remove the visible sparkle logo from AI-generated media. Invisible cryptographic benchmarks like SynthID remain intact for detection tools.
3. Apple partners with Alibaba to train custom AI model for China — Apple developed a China-focused large language model using Alibaba's infrastructure to navigate strict regulatory requirements. The collaboration highlights the necessary compromises for delivering global AI features.
4. OpenAI introduces Ultrafast inference tier for GPT-5.6 Sol — Leveraging a partnership with Cerebras, OpenAI's new tier delivers up to 750 output tokens per second. The move turns inference speed into a distinct commercial product.
5. Alibaba releases open weights for 27-billion parameter Qwen 3.8 — The highly capable dense model is licensed under Apache 2.0 and natively processes up to 262,000 tokens of context. It reportedly outperforms larger models in complex coding tasks.

Five stories. Five minutes. Every weekday.