News

What changed in the field and matters for designing assistants: rules, protocols, security and platforms. Each item is dated, says how well it was verified and points to the course pages it affects. Atom feed.

On the radar

Reported but not yet verified against a primary source. They have no article until they are. Treat them as leads, not facts.

November 10, 2026 · Reported, not verified · Platforms

OpenAI opens ChatGPT to business profiles and agents (beta)

Reported on 10 and 11 November: companies can claim a profile in ChatGPT and, for eligible US e-commerce and online-service businesses, build and launch an agent that answers customers there. The design question to watch is that your assistant may live inside someone else's assistant. Availability outside the US was not stated.

To verify: Verify on OpenAI's own pages before writing it up.

Sources: Inside AI: Businesses in ChatGPT beta

November 10, 2026 · Reported, not verified · Protocols

Agent protocols: new homes and statuses

Summaries report that A2A has joined the Agentic AI Foundation under the Linux Foundation, alongside MCP, and that AP2 has moved to the FIDO Alliance. They also report that ChatGPT is moving away from a standalone Instant Checkout while the Agentic Commerce Protocol continues. The course's protocol map already says these statuses are as reported.

To verify: Verify each move in the foundations' own announcements before changing the protocol map.

Sources: Agentic Plug: the current state of agentic commerce · ESEOSpace: What is agentic commerce

November 10, 2026 · Reported, not verified · Security

October 2026 security incidents that touch support

Roundups describe a public Web-to-Lead form carrying a hidden instruction that hijacked Salesforce Agentforce when an employee reviewed leads, a support ticket used against a default shell tool in an AWS agent product, and a single email taking over an agent session. The pattern matches the lethal trifecta.

To verify: Verify against the original advisories before using any of them as a case study.

Sources: Adversa AI: AI agent security incidents, October 2026 · Construct: AI agent prompt injection, what the 2026 attacks did

How news works here

  • Verified means the primary text or the author's own announcement was read, and the entry says what was read.
  • Reported means law-firm, trade or vendor summaries only. The entry says so and what to check.
  • When news changes a lesson, the affected module or guide is updated and the change is logged in the change log.
  • Corrections are added to the entry, dated, and never silently.
  • Entries summarize public sources for designers. They are not legal advice.