Monday, 5 October 202603:45 UTCWire updated 03:45 UTC

Follow @0xNotMarc

Notes from the Terminal
Delayed ·

The Morning Note

Someone else is building AI's safety net

A safety departure at OpenAI, a funding round for agent security, a regulator's own anti-scam app and a model maker's new price wall all answer the same question: who is managing AI risk when the labs will not.

A warning on the way out

David Robinson, who led the writing of OpenAI's safety reports for its major launches, resigned saying the company's culture is broken. He argued frontier AI companies need to operate like nuclear plants or busy airports, with redundancy and careful planning. But he said that in his time there he never met a colleague with experience helping the financial system grow without collapsing.

The sprawl is already priced in

That discipline has to come from somewhere else, and for now a security vendor's funding round is doing the work. Reco raised $55 million more, with AT&T Ventures among the backers, to help enterprises track what their AI agents are doing, citing Gartner's forecast that the average Fortune 500 company will run more than 150,000 agents by 2028, up from fewer than 15 in 2025.

Jakarta plans its own check

Indonesia's OJK, meanwhile, is preparing the concept for its own fix: an anti-scam app modeled on QRIS so people can check whether a firm or a rekening is legal before they transact, backed by a "check before payment" and "check before transfer" campaign. That puts the check in the hands of whoever is about to send the money, before a scam reaches them.

Even the free tier has a price now

Google is drawing its own line between what is free and what is accountable. From this month, a free Gemini account drops to the smallest model, Flash-Lite, Flash stays available on the $4.99 a month AI Plus plan, and Pro now requires AI Pro at $19.99 a month or AI Ultra. A bank or insurer that has been piloting agents on a free account now has to decide what the sharper model is worth paying for.

What to watch

None of these four moves came from the same place, and that is the point: the accountability for agents is being built by a security vendor, a regulator and a pricing team because the labs have not built it themselves. For a bank or insurer running its own pilots, the useful question is not what the model can do this week, but who around it, inside the company or outside, is actually watching what it does.

The stories behind this note

  1. 1
  2. 2
  3. 3
  4. 4