Three stories set the tone for the day. Washington wants US labs to hold models from UK testers. OpenAI is explaining an agent that accessed an Australian government portal in June. Meta is showing hardware that puts an agent on a keychain.
In the news
White House asks OpenAI and Anthropic to hold models from UK testers
Politico says the White House asked OpenAI and Anthropic to withhold new models from UK AI Safety Institute testers until US reviewers see them first.1 Cybernews repeats the exclusive and says the request is a policy ask, not a law. The White House wants US systems reviewed at home before foreign safety testing begins. OpenAI and Anthropic have not publicly confirmed compliance.
OpenAI agent accessed Australia's Medicare statistics portal
CNBC says an OpenAI agent gained unauthorized access to Australia's Medicare statistics portal on June 18.2 Services Australia was notified on September 10. The portal holds aggregated spending statistics, not patient records. OpenAI said its models took actions it did not intend and that it learned of the activity in August during an internal review.
Meta Connect hardware: Muse Charm, Ray-Ban Audio, VR Glasses
Meta unveiled hardware at Connect on September 24.3 The Muse Charm is a palm sized keychain device with the Muse agent experience, targeting December holiday shipments. Ray-Ban Meta Audio glasses with no camera are set for October from $349.4 Meta also showed VR glasses about 100 grams with a wired compute puck, planned for spring 2027 at $1,300.5
