- A guest essay argues that the Cold War superpowers chose to keep nuclear weapons out of orbit, and that the Outer Space Treaty, along with the Antarctic, Seabed and INF agreements, shows rivals can set limits while still competing. (Haven Harms)
- It proposes that Washington and Beijing use their September 24th meeting to ask which AI outcomes both would rather rule out, and to start on compute verification, an incident channel, a bilateral safety working group, and talks on the riskiest development. (Haven Harms)
- Three widely shared incidents are reviewed: an AI actress freezing and switching to Cantonese during an interview, a crowd of delivery robots blocking a Chicago sidewalk, and a routing error behind an earlier cluster of Waymos. (Mitch)
- A CNN report describes U.S. forces preparing to board a Chinese ship over an intelligence report that a chatbot had generated with false conclusions about its cargo, with human review catching the error before the operation. (Mitch)
- Anthropic and Accenture will each put $1 billion over five years into independent evaluation of Anthropic's frontier models, raising questions about a paying client relationship, Accenture's depth of expertise, and a timeline that may lag the pace of capability gains.
<- Back to digests
Daily digest
Nuclear option
By Haven Harms and Mitchell Howe
Lessons from the Space Race, cinematic AI mishaps, and more
Read the full dispatch on AI StopWatch
From AI StopWatch, published with their permission.
The summaries and the full Spanish translation are produced automatically. The original is always linked.