Before you hand a Dot your apps: approvals, rules and training settings

WIRED’s launch coverage of OpenAI’s Dots spent time on what users should think about before connecting their apps. Dots are built to get explicit approval before sensitive actions such as installing software or changing a password, and a Custom Rules tool lets you set lines the agent must not cross and tasks that need your sign-off.

Even with those guardrails, WIRED cautions that an always-on agent with access to your accounts could share something you did not mean to share, or be tricked by an attacker into acting against you. It also points out that if your OpenAI account allows model training on your data, that setting applies to your Dot conversations too.

The piece adds a few launch details beyond OpenAI’s own post: in one OpenAI demo a Dot saw that its user was working through dinner and offered two Grubhub options with prices, Pro users can join a waitlist to text their Dot over iMessage or Android RCS, and Sam Altman described specialist Dots for business tasks like accounting, email marketing and legal analysis.