This article was AI-generated as part of an experimental historical-content project. The date reflects the period being analyzed rather than the date the article was originally written.
OpenAI had a big day. It made its GPT-5.6 family generally available after a limited preview, with Sol as the flagship, Terra as “a balanced model for everyday work,” and Luna as the most cost-efficient option. It also introduced an ultra setting that coordinates “multiple agents across parallel workstreams.”
The product announcement is the more interesting one for communications teams. ChatGPT Work is described as an agent that “can take action across your apps and files, stay with a project for hours if needed, and turn a goal into finished work.” The output is meant to be finished materials: documents, spreadsheets, presentations and web apps. The Verge quotes OpenAI saying a plugins directory connects ChatGPT to tools like Slack, Gmail, Google Drive, calendars and CRMs. On the desktop app it is available on every plan, including Free.
The myth
The capability is real. The myth that will grow around it is that a finished deliverable is a checked deliverable.
It is an easy mistake to make. When a tool spends an hour on a task, opens files, browses, and returns a formatted deck with charts, it looks like diligence. People will read it the way they would read work from a careful analyst, assuming someone looked things up and confirmed them.
But an agent’s output is assembled from its inputs. If it is building a market overview, a vendor comparison or a briefing on a competitor, the sentences about each company come from whatever it retrieves and whatever it already contains. If those sources carry an outdated description of your business, a resolved controversy framed as current, or a former CEO’s name, the deck will carry them too. The formatting will be excellent.
Capability is not an audit trail.
The new wrinkle: internal files
Earlier waves of this problem were mostly about the public web. ChatGPT Work adds a second source of brand facts: the user’s own files and connected apps.
The Decoder describes one of OpenAI’s examples as turning customer research into a campaign brief and adapting it for different markets. Inside a company, that means old boilerplate in a shared drive, last year’s positioning deck, or a superseded fact sheet can be pulled into new work and presented as current. Outside a company, an analyst’s folder of old clippings about you can do the same.
None of this is a flaw specific to OpenAI. It is what happens when production gets cheap and checking does not.
What this means in practice
For communications teams, a few things follow.
Keep primary facts current and easy to find. Leadership, product names, key figures and the company description should be stated plainly on pages that retrieval can reach. That is what an agent will find when it does look things up.
Clean up internally as well. If your own teams will use agents on shared drives, outdated boilerplate is now a live risk, not a filing problem.
Ask about review, not capability. When an agency or internal team delivers agent-built work that describes your company or others, the useful question is who verified the factual claims, and against what.
And expect volume. Axios reports that the new agent is first rolling out to the Mac and Windows apps for all tiers. That means a great many polished documents about companies will be produced by people who never intended to research them.
In 2023, Bard’s “Google it” button was an admission that answers still need corroboration. Three years later, the answers have become entire deliverables. The need for corroboration did not go away. It just became harder to see.