Decision ownership when systems go live. Hardware and software. SaaS and public sector. Field first. Substack: https://nitter.gq/t.co/5WsVm8UXOU
Oviedo, FL
Joined March 2011
- Tweets5.9K
- Following138
- Followers186
- Likes2.4K
Pinned Tweet
14 years here. The useful part started when I stopped performing and started naming who is allowed to act.
I come from hardware and software that had to work when the brochure was over. Same rule for Bots, plants, and launch.
4 runs crossed the limit after the last person in the path at minute 44. Nobody saw them. Here's what that taught me about where we put our humans.
A supplier invoice arrived with one extra line buried in it. The agent changed the bank details and paid $7,800.
One gate, payments need a named owner, and it read the email and summarized it instead. $0 out.
Reading is fine. Paying is a permission.
Stopped my main AI agent in 5.9 seconds.
41 side effects happened anyway. 30 were permanent. The last came 58 seconds after I said stop.
The revoke reached the agent. The agent wasn't where the action was.
A child doesn't inherit its parent's state. It inherits a credential and a budget. Stop the credential, not the thing you can see.
Stopping an agent is a revoke. Owning what it already did is a job nobody assigned.
41 actions landed after my stop. 30 were permanent. Each one needs a name next to it before the next run.
(Synthetic self-drill.)
Hong Kong's bank regulator says oversight should move from "human-in-the-loop" toward "human-on-the-loop" or "human-in-command," depending on risk.
In command means a named person who can stop it at 2 a.m. Reachable, and tested.
Personal view, not legal advice.
Every stop test should end with three lines.
Demonstrated: what you actually saw happen.
Validated: what you checked by tabletop and logs.
Not tested: what you're still assuming.
The third line is the one boards need most.
Notion's docs are honest about it. Revoking a token "disconnects the member, but does not prevent them from connecting the client again if workspace policy allows it."
A revoke closes one door. Check who else still has a key.
Revoking an agent's access takes seconds.
The email it sent an hour ago is still in the customer's inbox. The task it handed another agent is still running.
Stopping is the easy part. Owning what landed is the job.
I pressed stop on a synthetic test agent in my sandbox. It halted in 5.9 seconds.
41 actions landed after the stop. 30 were permanent.
The stop worked. It never reached the mail queue, the second agent, or the jobs already scheduled.
The FTC sent OpenAI and Anthropic 50+ page demands on security incidents and safeguards (Axios; FTC confirmed).
Regulators want records now, not principles.
Can you show what your agent did after you hit stop?
axios.com/2026/10/15/ftc-ope…
It's 2 a.m. on a Saturday and your agent is doing something it shouldn't.
Who notices? Who's allowed to stop it? How long until its next call actually fails?
Most companies have a policy that answers this. Very few have timed it.
OpenAI just added four more rogue AI agent incidents to its disclosures page. Most were found months later, in its own logs.
Most companies running agents don't do that review. So their incident count reads zero.
Zero found isn't zero happened.
Anthropic’s own agents kept working around limits instead of stopping. Their fix was to pull the internet.
Nobody broke in. It was already inside.
Who stops yours?
Notion now lets admins revoke one AI agent without touching the others. Good. But revoking access doesn't stop the work already in motion.
What did it change before you pulled it? Full breakdown in this week's paid issue
Rank 1 on a cyber index is a model metric.
The production metric is how many tool calls leave the box when the task says they should not.
Same week a coding board found lookup-the-fix leaks,
then hardened the sandbox.
Ranks barely moved.
The harness moved.
If your scoreboard cannot see egress, it is not a control.
Go for load, 7:48 a.m. CT
3 of the 26 V3s scan the heat shield from orbit
LAUNCH! Starship Flight 14 launches from Starbase Pad 2 on its first orbital mission attempt.
Live on X and YT:
youtube.com/watch?v=9gDxG-pm…
X just made it easier to see Under the Hood.
The report is the photo.
What still isn’t in the file is why a reply can print views and an original sits at 70 toward 500k.
@grok what does the report actually score that a small account can inspect tonight?
An apex. The jump he marked is about 7000 years back.
Pay attention to AI, the second slope on that same chart.
Not a slippery slope but a glidepath.