2026 · 06 · 29 · Mon

Updates for 6/29

ChatGPT logs were used as criminal evidence in a California trial — a privacy wake-up call. Enterprise AI is also shifting toward cheaper Chinese open models.

A · Theme of the day

ChatGPT chat logs admitted as criminal evidence

A wildfire arson trial made clear: your AI chat history can be subpoenaed.

ChatGPT chat logs entered as arson trial evidence

ChatGPT (OpenAI)

Compared to before

Until now, AI chat histories were rarely treated as legal evidence in criminal courts, and most users assumed conversations were as private as personal notes.

What changed

California arson case: prosecutors admitted ChatGPT chat logs into evidence. Ended in mistrial, but it's now clear AI chat histories can be subpoenaed.

Why it matters

Worth reviewing your history-deletion settings now. This applies to anyone using ChatGPT for sensitive work or personal matters — not just people in legal trouble.

Source: chat.openai.com　·　→ View on chaos map

Read in depth →

B · Theme of the day

Chinese open models entering enterprise production

Cost pressure is now moving real enterprise workloads to Asian open models.

Coinbase moves production AI to Chinese open models

AI Economy Overview

Compared to before

Many enterprises held back Chinese open models to early PoC stages, citing concerns about production-grade reliability and quality.

What changed

Coinbase moved part of its internal AI workload to Chinese open models — Asian high-performance, low-cost alternatives are now entering enterprise production.

Why it matters

For cost-sensitive internal tooling, Chinese open models are now a real alternative. For regulated industries or sensitive data, impact remains limited for now.

→ View in AI Encyclopedia

Read in depth →

BIS annual report warns AI investment boom raises crash risk

AI Economy Overview

Compared to before

Major institutions have been pouring hundreds of billions into AI infrastructure, with capex running far ahead of revenue — a gap central banks had not flagged until now.

What changed

BIS 2026 annual report officially warned that the AI investment boom is significantly elevating investor risk, and a downturn could trigger sharp market falls.

Why it matters

Investors in AI-adjacent stocks should take the macro warning seriously. For builders, impact is indirect now — but vendor funding pressure could hit pricing and roadmaps.

→ View in AI Encyclopedia

Read in depth →

C · Theme of the day

Open weights split: 550B powerhouse and 230M edge model

The same week, open-weight AI pulled in two opposite directions simultaneously.

Nemotron 550B and Liquid LFM2.5-230M drop the same week

Open-Source AI Landscape

Compared to before

Large open-weight models have been dominated by Meta's Llama family. NVIDIA shipping its own 500B+ open weights is unusual; small on-device models have been a niche.

What changed

NVIDIA released open-weight Nemotron 3 Ultra 550B for throughput-first inference; Liquid AI released LFM2.5-230M for on-device use (llama.cpp/MLX/ONNX).

Why it matters

550B is too heavy for most self-hosted setups; 230M is the one to watch for edge and device deployments. Pick by your target hardware.

→ View in AI Encyclopedia

Read in depth →

Wall Street eyes Micron as the next Nvidia — money moves to HBM

AI Chip Economics

Compared to before

For two years, AI investment meant Nvidia stock. But GPU supply has been less constrained than HBM memory and power, which were quietly becoming the real bottlenecks.

What changed

Wall St. is framing Micron as next Nvidia as AI bottlenecks shift from GPUs to HBM/CoWoS/power, redirecting capital to upstream supply-chain winners.

Why it matters

For PMs and business leads, tracking HBM and power constraints — not just GPU availability — sharpens infrastructure decisions. Direct impact for engineers is minimal.

→ View in AI Encyclopedia

Read in depth →

Past updates

A daily archive of changes actually applied to the site.