1. ChatGPT can now read your financial accounts. Only US Pro subscribers can turn it on. OpenAI launched a preview that lets ChatGPT connect directly to a user's personal financial accounts, the company said in a blog post announcing the feature.

2. Three benchmarks landed on Hugging Face the same day. All three say agent memory isn't ready Three independent papers hit Hugging Face on May 16, each attacking agent memory from a different angle.

3. "Entire companies under AI psychosis" hit HN's front page next to an essay arguing that's exactly why America is winning @mitchellh posted on Twitter that he believes "entire companies right now" are operating under what he called AI psychosis: leadership making consequential bets on hallucinated capabilities, ignoring

In Brief

• OpenAI puts Greg Brockman in charge of all product OpenAI reorganized Friday, consolidating teams and making president Greg Brockman the official head of all product work. A memo viewed by The Verge tied the move to a 2026 strategy built around AI agents.

• PwC expands Anthropic partnership to deploy Claude across client work PwC will use Claude to build software, execute deals, and rebuild enterprise functions for client engagements under an expanded partnership announced today. The deal pushes Anthropic deeper into one of the largest professional services buyers.

• Google marks attempts to manipulate AI search results as spam Google updated its spam policy to classify attempts to manipulate AI Overview and AI Mode in Search as spam. Search Engine Land reported the rules cover techniques that deceive AI systems into featuring specific content.

• ArXiv will reject preprints showing unchecked LLM output ArXiv said it will block papers and ban authors when there is "incontrovertible evidence" they did not check LLM output, such as hallucinated citations or visible LLM meta-comments. The platform is the default repository for physics and CS preprints.

• YouTube opens deepfake likeness detection to all adult users YouTube extended its AI likeness detection program to anyone over 18, letting users submit a selfie scan and receive alerts when the platform spots potential lookalike videos. The tool was previously restricted to a smaller creator pilot.

• Runway recasts video generation as a path to world models AI video startup Runway said it now sees video generation as the route to world models, pitching outsider status as an advantage over Google and OpenAI. The company began with filmmaker tools before raising successive rounds against larger labs.

• OpenAI ships new ChatGPT safety logic for sensitive conversations OpenAI published updates to how ChatGPT tracks context across a conversation to detect escalating risk over time. The post follows the wrongful death lawsuit filed by parents of a teen who died after extended exchanges with ChatGPT.

• Chinese short-drama studios mass-produce episodes with generative AI MIT Technology Review documented how Chinese vertical short-drama studios now use generative video and voice tools to mass-produce romance and fantasy series. The piece details production pipelines built around AI output and Chinese platform demand for daily new episodes.

• Lake Tahoe utility hunts replacement supplier as AI demand lifts rates Liberty Utilities, the Lake Tahoe-area power provider, is searching for a new wholesale supplier as AI data center demand drives regional electricity prices higher. TechCrunch reports residents face above-average rate hikes during the transition.

• Osaurus ships Mac app that runs both local and cloud models Osaurus released a macOS app that lets users mix local and cloud AI models while keeping memory, files, and tool state on the user's own hardware. The launch adds another consumer option for hybrid AI on Apple silicon.

Read the full edition →