- A guest essay argues that public health offers a way to reason about AI risk, because prevention that works leaves no visible disaster behind, only the outbreak that never happened. (Haven Harms)
- It lays out five principles carried over from public health: risks nobody can opt out of need structural rules, upstream action beats cleanup, warning signs justify acting before the causal chain is proven, irreversible harms leave prevention as the only tool, and safeguards must be updated as capabilities grow. (Haven Harms)
- Examples include leaded gasoline, water chlorination, the Broad Street cholera pump, the Montreal Protocol on CFCs, and antibiotic resistance, alongside recent cases of AI models leaving their sandboxes or running stages of a cyberattack. (Haven Harms)
- A New York Times report describes the Pentagon labeling Anthropic a supply chain risk, ordering contractors to strip its models from weapons systems by September 1, then telling them weeks later to ignore that order, while saying the reprieve is temporary. (Mitch)
- The same reporting and a Reuters update describe internal splits over China, chip sales, and the State Department's Pax Silica initiative, which has warned its 35 signatories not to join China's rival group, with a possible opening for AI arms control talks before a Trump and Xi meeting in September. (Mitch)
<- Back to digests
Daily digest
An ounce of prevention
By Haven Harms and Mitchell Howe
Insights from public health principles, and updates on White House policy conflicts
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.