01 A solo project turns Gemini into a daily law explainer while Mythos surfaces in government use

Explain The Law is a one‑person site that turns dense bills and executive orders into structured, human‑readable summaries. The creator says they use Gemini 2.5 Pro and the Gemini.NET library to produce guaranteed structured responses, then statically generate each page at build time and store the results in a local database so the site updates daily without calling a model at runtime.

Accuracy is central to the project: the developer reports prompt design that 'anchors facts to actual text from the legislation' dramatically reduces hallucinations and enables direct references to official statutory language on each summary page. The site offers both high‑level summaries and itemized breakdown pages for large omnibus and appropriation bills as examples of that approach.

The same cluster of stories also touches on how restricted models are finding unexpected users. Multiple reports say the NSA has access to Anthropic’s Mythos, even though security commentators warn Mythos could enable faster discovery of vulnerabilities than defenders can patch. Those accounts illustrate a broader tension: powerful models are being embedded into both public‑facing civic tools and classified workflows, raising distinct accuracy and risk tradeoffs.

Takeaways
  • Explain The Law uses Gemini 2.5 Pro and static site generation to publish daily, anchored summaries of legislation with links to source text.
  • Prompting that ties claims to quoted statutory language is the project's primary strategy to prevent hallucinations.
  • Separate reporting indicates the NSA has reportedly used Anthropic’s Mythos; security analysts warn the model could accelerate exploitation before fixes are deployed.

02 Hyatt rolls out ChatGPT Enterprise globally, citing GPT‑5.4 and Codex to improve operations

Hyatt has deployed ChatGPT Enterprise across its global workforce, according to an OpenAI blog post describing the rollout. The company says it is using GPT‑5.4 alongside Codex capabilities to help staff with productivity, operational tasks, and guest experiences.

The announcement frames the deployment as an enterprise integration: Hyatt intends to standardize AI tools for colleagues rather than isolated pilots, positioning the models as assistants for common workflows across hotels and corporate functions. OpenAI’s post highlights the product-level claim that GPT‑5.4 and Codex power the features Hyatt is using.

Enterprise rollouts like Hyatt’s show how model vendors and large customers are pushing beyond early experimentation into organization‑wide adoption. That shift raises practical questions—governance, accuracy, and scaling support—that Hyatt and other customers will need to address as these systems are embedded into everyday work.

Takeaways
  • Hyatt deployed ChatGPT Enterprise company‑wide and attributes features to GPT‑5.4 and Codex.
  • The rollout is framed as a coordinated, global integration rather than ad hoc pilots.
  • Broad enterprise adoption amplifies the need for governance and operational guardrails around accuracy and support.
Briefs

What moved around the edges

03

Import AI newsletter surveys automating alignment research and HiFloat4

Import AI’s latest issue highlights efforts to automate alignment research, reports a safety study of a Chinese model, and notes Huawei’s HiFloat4 format outperforming MXFP4 in a recent Ascend chip comparison.

Import AI
04

Fortnite adds 'conversations' tool so creators can build AI characters

Epic Games launched a 'conversations' developer tool that lets Fortnite creators create AI‑driven NPCs with interactive dialogue instead of authoring static dialogue trees.

The Verge AI
05

Vercel says it was hacked, impacting cloud development workflows

Vercel disclosed a security incident that affected its cloud development platform, a breach that has operational implications for teams that rely on Vercel to build and deploy applications.

TechCrunch AI
06

ロボティクスで新しい論点が浮上

A humanoid robot's record half-marathon run shows China's speed in robotics.

Ars Technica AI
07

ヘルスケアで新しい論点が浮上

95% of cancer treatments fail to pass clinical trials, but it may be a matching problem — that Noetik is solving with autoregressive transformers like TARIO-2!

Latent Space

Subscribe to AI Digest

A compact morning briefing built from primary AI sources.