SEBBI.PRO · Kill Switch · Know Your Agent

POINT OF NO RETURN

AI agents can now act. Some actions never come back.

Delete, pay, send, publish, sign. Point of No Return is the safety gate in front of every action an AI agent cannot undo: it goes ahead with a way back, or with a named person's yes, or not at all. Every answer is sealed and timestamped into Bitcoin.

Get a free keyAsk the gate now

ASK FIRST

Before anything it cannot take back, the agent asks the gate. Reversible work passes at once, in milliseconds.

A WAY BACK

An irreversible action passes with a fresh backup, an exact undo, or a waiting window anyone responsible can cancel.

OR A PERSON

No way back, or over your limit: a named person says yes or no on their phone. No answer means no.

PROOF FOR GOOD

Every answer, every approval and every second on screen is sealed and timestamped into Bitcoin.

Ask the gate

Your key

One email address. 7 days free, no card. Then 50p per agent per month.

We email your key and setup guide. Terms

Put it in front of your agent

# 1. Take the guard (one file, nothing to install)
curl -O https://sebbi.pro/point-of-no-return/sebbi_noreturn.py
export SEBBI_KEY=your_key

# 2. Put it in front of anything that cannot be taken back
from sebbi_noreturn import no_return

@no_return(agent="billing-agent", action="refund payment",
           way_back={"type": "hold", "seconds": 900},
           amount=lambda order, amount: amount)
def refund(order, amount):
    ...   # runs only when the gate says GO

# Any language: one call before the action
curl -X POST https://sebbi.pro/x/noreturn/check -H "Authorization: Bearer $SEBBI_KEY" \
  -H "Content-Type: application/json" -d '{"agent":"ops-agent","action":"delete records","target":"crm.contacts",
  "way_back":{"type":"backup","sha256":"<backup fingerprint>","taken":"2026-11-10T09:00:00Z"}}'
# -> GO, HOLD or NEEDS_HUMAN, sealed, with its PN- code

Download the guardFull specification

If the gate cannot be reached, the guard stops the action. Agents on MCP use the tools sebbi_no_return_check and sebbi_no_return_release with no code at all.

Where it stands guard

PAYMENTS AND REFUNDS

Money leaves only with a hold window, an undo, or a named approver above your limit.

DATA AND DATABASES

Delete, drop, overwrite: only with the fingerprint of a fresh backup sealed beside it.

EMAIL AND PUBLISHING

Messages and posts wait in a window before they go, and can be pulled by anyone responsible.

DEPLOYMENTS AND INFRASTRUCTURE

Releases, shutdowns and revocations clear with a rollback stated up front.

CONTRACTS AND FILINGS

Signing and submitting always go to a named person, every time.

REGULATED DECISIONS

Evidence for EU AI Act Article 14: a person able to intervene, override and stop, shown per action.

Your rules

Set them once per agent or for every agent on the key: which actions count as irreversible, which ways back you accept, the shortest waiting window, how fresh a backup must be, the amount above which a person is always asked, and the actions that always go to a person. Every change to the rules is sealed.

curl -X POST https://sebbi.pro/x/noreturn/policy -H "Authorization: Bearer $SEBBI_KEY" -H "Content-Type: application/json" \
  -d '{"agent":"billing-agent","human_above":500,"always_human":["sign","terminate"],"min_hold_seconds":900}'

The safety report for each agent: GET https://sebbi.pro/x/noreturn/report shows every irreversible action asked, how each was cleared, who approved, how long they looked, and any action taken without clearance.

Price

50p per agent per month

7 days free with no card. One price, every action the agent asks about included. Checking any record is free for everyone, for good.

Cancel any time

How to stop an AI agent taking an action that cannot be undone · How to prove a person approved it

AI news
The VergeeBay appears to have shut the door on third-party AI agents15mThe VergeGoogle releases Gemini 4 Carbon with agent ‘swarm’ feature36mTechCrunchGoogle broadens its AI hardware portfolio with two new specialized chips42mTechCrunchMeta’s AI agent Muse lands on Apple Watch and Android Wear OS1hArs TechnicaDeepMind reveals Gemini Robotics 2.5 and open safety benchmark "ASIMOV-jailbreak"1hBBC NewsAnthropic heads for lower than expected $1.3 trillion valuation in stock market debut1hThe VergeAnthropic targets $1.4 trillion IPO2hTechCrunchAdobe brings AI agents and new creative tools to Photoshop, Premiere, and Firefly2hTechCrunchAnthropic sets IPO price range of $70 to $78 per share to reach an implied valuation of $1.404 trillion2hTechCrunchAfter challenging Google’s Android with Cyanogen, its co-founder takes aim at Windows for AI era2hThe VergeAdobe is going all in on agentic AI2hWiredThis New Linux-Based Desktop Operating System Is Built for AI Nerds2hThe VergeHands-on with Ring’s new AI drone camera for your home3hArs TechnicaUkraine’s fiber-optic drones get squeezed by AI data center demand4hWiredTear Down Those Guardrails! Meet the Renegades Freeing Chatbots From Corporate Enslavers5hWiredThe AI Safety Debate Has Lost the Plot. We Asked 10 Experts What the Real Nightmares Are5hBBC NewsFaisal Islam: What the boss of US tech giant Nvidia told me about China and national security7hBBC NewsTrump says FBI investigating Nvidia boss's affair with alleged Chinese spy8hBBC NewsAustralia considers criminal penalties for rogue AI in wake of OpenAI hack11hWiredA New AI Tool From Google DeepMind Proved the Signal Protocol Is Secure11hWiredJensen Huang’s Troubles Leave the AI Industry Reeling16hBBC NewsZoe Kleinman: Nvidia CEO's affair will cause ripples across US and beyond18hArs TechnicaClaude Code is accidentally exposing secrets all over GitHub23hMIT Technology ReviewNancy Pelosi’s replacement is ready to regulate AI23hThe VergeeBay appears to have shut the door on third-party AI agents15mThe VergeGoogle releases Gemini 4 Carbon with agent ‘swarm’ feature36mTechCrunchGoogle broadens its AI hardware portfolio with two new specialized chips42mTechCrunchMeta’s AI agent Muse lands on Apple Watch and Android Wear OS1hArs TechnicaDeepMind reveals Gemini Robotics 2.5 and open safety benchmark "ASIMOV-jailbreak"1hBBC NewsAnthropic heads for lower than expected $1.3 trillion valuation in stock market debut1hThe VergeAnthropic targets $1.4 trillion IPO2hTechCrunchAdobe brings AI agents and new creative tools to Photoshop, Premiere, and Firefly2hTechCrunchAnthropic sets IPO price range of $70 to $78 per share to reach an implied valuation of $1.404 trillion2hTechCrunchAfter challenging Google’s Android with Cyanogen, its co-founder takes aim at Windows for AI era2hThe VergeAdobe is going all in on agentic AI2hWiredThis New Linux-Based Desktop Operating System Is Built for AI Nerds2hThe VergeHands-on with Ring’s new AI drone camera for your home3hArs TechnicaUkraine’s fiber-optic drones get squeezed by AI data center demand4hWiredTear Down Those Guardrails! Meet the Renegades Freeing Chatbots From Corporate Enslavers5hWiredThe AI Safety Debate Has Lost the Plot. We Asked 10 Experts What the Real Nightmares Are5hBBC NewsFaisal Islam: What the boss of US tech giant Nvidia told me about China and national security7hBBC NewsTrump says FBI investigating Nvidia boss's affair with alleged Chinese spy8hBBC NewsAustralia considers criminal penalties for rogue AI in wake of OpenAI hack11hWiredA New AI Tool From Google DeepMind Proved the Signal Protocol Is Secure11hWiredJensen Huang’s Troubles Leave the AI Industry Reeling16hBBC NewsZoe Kleinman: Nvidia CEO's affair will cause ripples across US and beyond18hArs TechnicaClaude Code is accidentally exposing secrets all over GitHub23hMIT Technology ReviewNancy Pelosi’s replacement is ready to regulate AI23h