AI News, Aug 15, 2026: Nvidia reduces OpenAI backing as Anthropic reports revenue surge

Today's news highlights the real-world friction of AI adoption, from data privacy disputes on Twitch to hidden prompt injections in court.

1. 01
   Anthropic explains how Claude text watermarking will workTechCrunch

   Anthropic shared technical details on its upcoming watermarking system, which subtly adjusts word choices to identify text generated by its Claude models.

   Why it matters

   Invisible watermarks help platforms detect AI-written text, which is crucial for combatting academic cheating and automated spam.
2. 02
   SpaceX completes acquisition of AI coding startup CursorTechCrunch

   SpaceX has officially finalized its acquisition of Cursor, a prominent startup building AI-powered code editors.

   Why it matters

   The buyout brings one of the most popular AI-assisted programming tools under the control of Elon Musk's aerospace company.
3. 03
   Nvidia reduces OpenAI backing as Anthropic reports revenue surgeThe Decoder

   Nvidia nearly halved its financial guarantee for OpenAI's Ohio data center to $120 billion, while competitor Anthropic reported its revenue surged to $11.5 billion.

   Why it matters

   Shifts in tech investments could change which AI tools survive, while Anthropic's growth suggests enterprise demand for AI remains strong.
4. 04
   AI-generated books flood Amazon and lower earnings for human authorsThe Decoder

   A new study found that AI-generated books now make up 20 percent of Amazon's self-published catalog, driving down average revenues for human authors.

   Why it matters

   The surge of low-quality AI books is diluting self-publishing platforms, making it harder for human writers to make a living.
5. 05
   Amazon uses Twitch streams to train AI unless creators opt outWIRED AI

   Amazon-owned Twitch announced that creators can opt out of having their broadcast content used for AI training, sparking backlash over default data collection.

   Why it matters

   Streamers must now manually change their settings to prevent their live video and audio content from being used to build Amazon's AI models.
6. 06
   Plaintiff uses hidden text in court filings to trick automated AI reviewsThe Decoder

   A Connecticut plaintiff had electronic filing privileges revoked after embedding invisible, white-on-white text prompts in court documents to influence AI readers.

   Why it matters

   This real-world exploit shows how easily automated legal and administrative screening systems can be manipulated by hidden instructions.
7. 07
   World Labs releases simulation engine to train robots in virtual worldsThe Decoder

   AI pioneer Fei-Fei Li's startup, World Labs, unveiled a system that creates thousands of simulated variations of a single physical task to train robot controllers.

   Why it matters

   Generating thousands of virtual training scenarios from a single real-world task could dramatically speed up how quickly physical robots learn.