Newsletter

Five minutes. What happened, and what to do about it.

HIGHLIGHT

ChatGPT dots: always-on agents that keep working when you leave

Set a goal, close the chat. The dot carries on in the background.

TECHCRUNCH|29 SEP

Anthropic: a new open model can build cyber attacks with few safeguards

Anthropic· 29 Sep

Anthropic tested GLM-5.3, a model from Zhipu AI. It built working end-to-end exploits in 12% of attempts, close to Claude Mythos Preview's 14%. Simple tricks got past its protections 64% to 100% of the time. A researcher built an exploit for a known flaw in about 20 minutes of human effort and roughly $20.40 of compute.

  • Install security updates within days, not months. Cheap AI help means known flaws get attacked faster.
  • Turn on automatic updates for routers, websites and plugins, and list anything that cannot update itself.

Source: Anthropic. This is our short summary of their reporting. The commentary in it and the things to do are SeedFoundry's opinion. Picture from the source, unless it is our own graphic. How we source and credit

That is tue 29 sep, all of it.

These are short summaries of other people's reporting, with our own commentary. Each story names its source and links to the original. How we source and credit