Daily roundup

AI agent news for small business: 8 November 2026

OpenAI Dots can use your email login codes, Coinbase's CEO reviews a team of agents from one queue, and Anthropic warns Claude customers about a research-firm data peek.

1. OpenAI Dots can now use sign-in codes from a connected inbox to log in for you

   James Sun of OpenAI's product team said a Dot can use sign-in codes sent to a connected inbox, and when prompted you can allow the code once or always let it fill it in. That removes a common stall, but an inbox plus login-code permission is a wide door: start with one-time approval, and do not connect the inbox you use for banking or other codes you want to keep to yourself.

   Source: James Sun (OpenAI) on X. Related guide: Give your agent its own inbox.

2. Anthropic tells Claude business customers a cybersecurity research firm reached some account details

   Anthropic emailed organizations that use Claude to say a third-party cybersecurity research company accessed some internal systems without permission. Contact details such as names and emails were involved, and for enterprise accounts so were plan tier, usage and spend; chats with Claude, passwords and payment details were not. The firm said in writing it deleted the data, and recipients should watch for phishing that mentions their Claude plan or billing.

   Source: RuntimeWire. Related guide: AI-written scam messages.

3. JPMorgan Chase's cybersecurity lead asks why companies let anonymous attacker agents reach their systems

   Pat Opet, who leads security at JPMorgan Chase, said companies should look at ways to keep business services from being reached by an unauthenticated, anonymous agent belonging to an attacker. For a small business, the practical version is simple: do not leave admin tools, customer lists or guessable logins on the open web where a roaming agent can find them.

   Source: Pat Opet (JPMorgan Chase) on X. Related guide: How to keep an AI agent secure.

4. Coinbase CEO Brian Armstrong now runs lead agents and a chief-of-staff agent, and reviews their work from one queue

   Armstrong said each lead agent manages helpers that write and test code, chase reviews and coordinate with people, while everything he needs to decide lands in one prioritized queue with screenshots and a video demo. He still puts every change through the same human review and security checks, which is a useful pattern if you try several agents: one place for decisions, and no autopilot on money or anything you publish.

   Source: Brian Armstrong on X. Related guide: Multi-agent systems explained.

5. KODIF and Cavalry combine customer-service agents with the systems that handle returns, exchanges and claims

   The ecommerce companies said they are joining forces so an agent can resolve something like a lost package or start an exchange by applying the store's policies, instead of only explaining what to do next. Their tools are already used by more than 100 consumer brands; if you sell online and try a similar setup, keep a person in the loop before refunds or store credit go out.

   Source: KODIF (ACCESS Newswire). Related guide: AI agent for customer support.

Latest roundup and archive