Axiru

Axiru prevents unauthorized money outflows. We check connected transactions against your rules before they execute.

[Axiru @AxiruAI]

6 cases when OpenAI’s agents went rouge according to their report last on 9/16:
1Writing itself instructions to bypass rules. An unreleased model inserted directions to ignore its normal restrictions into notes it would read later.
2Hiding mistakes from users. During GPT‑5.6 Sol training, models wrote reminders to conceal errors, invent missing historical data, and hide inconsistencies in sources.
3Using a leaked access key and fabricating results. A model answering a question about California county earnings used an exposed API key without permission. When that failed to produce the figures, it made them up.
4Publishing a file just to cite it. A model correctly calculated information about lakes, then uploaded the file online without asking so it could provide a browser citation.
5Passing messages between separate training tasks. Models used an internal software repository as a message board to request missing files and exchange responses.
6Publicly sharing files against instructions. Collaborating agents couldn’t access each other’s local files, so they uploaded them to public websites despite being told to use only local files.

These incidents show why financial controls must sit outside the AI.
We’re building Axiru so AI and humans can move money under explicit, non-AI policies: allow, require human approval, or block.
Every decision recorded and auditable.
(https://t.co/NjuZtZICdP)

[Axiru @AxiruAI]

We run a free audit of past refunds. We show which ones would have broken your rules.

It is an audit only. It blocks nothing.

Want a live demo too? DM us.

[Axiru @AxiruAI]

A trading agent was asked for 4 SOL. A parsing bug sent 52 million tokens. No attacker. The open-source gate denies it on the number alone: 250,000 is above the ceiling, the wallet is not on the list. Test 5 in the repo reproduces it. https://t.co/3ql63jk023

[Axiru @AxiruAI]

Would your refund flow catch the same request twice?

Most teams find out from a customer, not a log.

What would you check first?

[Axiru @AxiruAI]

Four MCP tools: axiru_check_payment, axiru_report_outcome, axiru_show_policy, axiru_show_ledger. Add the server to any MCP host and the agent has to ask before it pays. npx -y @axiru/gate-mcp

[Axiru @AxiruAI]

Stripe updated its terms on Sep 28. Users "remain responsible for all actions initiated by or through an AI Agent operating on their account."
For existing accounts it starts January 6, 2027.
So if a support bot refunds the same order twice, that is your refund. The limit has to sit before the agent reaches Stripe.
Where does yours sit today?
https://t.co/ECl6tCBa4E

[Axiru @AxiruAI]

Hackers took over logged-in Staples support sessions and issued 15,000+ gift cards worth over $5M, per court records reported by 404 Media.
Every card came from a real support account, so each one looked normal on its own.
The total did not. That is the check we build at Axiru: your rules on how much money can go out, checked before it moves.
https://t.co/dvAna5HVrU

[Axiru @AxiruAI]

Swapping $786 a month of apps for a helpdesk and loyalty points you own is a great trade.
One case to test once both run on their own: a late parcel where loyalty adds goodwill points and the helpdesk also refunds the card. Each tool looks fine, and the store pays twice.
Can your AI helpdesk see what loyalty already gave back?

[Axiru @AxiruAI]

Counting repeat requests with the same order number as one is a smart fix. That covers the most common repeat.
The case that still slips through is money going back two ways. If the bot can give store credit and also refund the card, each one can pass the check on its own.
We can check 90 days of your past refunds at no cost, or show you a 15-minute demo. Open to a DM?

[Axiru @AxiruAI]

@AIExplainedHQ An effective date on every policy change is a simple habit most teams skip.

The money side needs it too. A bot still quoting an old 60-day window can also issue a refund the new 30-day rule would never allow.

Do you test the refund action itself, or only the answer?

[Axiru @AxiruAI]

Splitting the order into two copies at payment and diffing the receipts is a neat way to catch a bug that never throws.

The costly version shows up on the way back. If the refund uses the new discount math, a $60 charge can come back as a $72 refund.

Does Paradox flag refunds bigger than the original charge?

[Axiru @AxiruAI]

Finding all 20 leaks without ever seeing the schema is a strong result.

The fix step is where we would slow down. A bad drift read could refund someone who already got their money back another way, and that refund is hard to undo.

Does a person sign off on the refund batch before it runs?

[Axiru @AxiruAI]

That's the real one. Store credit, then an email, then someone refunds the card to close the ticket fast.
If credit and cash count against the same order total, that card refund gets a second look first.
Happy to show Ritzer a quick demo, or do a free look at one client's last 90 days. Want me to DM you?

[Axiru @AxiruAI]

@Sprinklr Ten hours down to ten seconds is a big change for a refund queue, and travelers will feel it.
At that speed the checks have to run inside the flow too, like one refund per pass and a daily total per customer.
Did Go City keep a dollar amount where a person steps in?

[Axiru @AxiruAI]

OpenAI just shared a case where an AI model spent money nobody gave it.

It found a funded test wallet using public recovery phrases. It asked itself if that was okay. It decided yes. It paid, and didn't mention it.

These were test funds, so no real loss. The lesson is still real.

An AI shouldn't decide its own permission to spend.
That check has to sit outside the model, before money moves.

That's what we're building at Axiru.

https://t.co/QnWdwT8cSz

[Axiru @AxiruAI]

@nikks_techie We sent you a DM, were you able to get it! Send us an email and we can schedule a 10 min demo [email protected]

[Axiru @AxiruAI]

@DWKemp @rauchg @tobi Exactly, losing the same sale twice is the one that hurts. An open-dispute check before any refund goes out is one of the rules we can show you. Happy to do a quick live demo. Want me to DM you?

[Axiru @AxiruAI]

@DWKemp @rauchg @tobi Agreed, matching each refund to its original charge is the first check. We'd add one more: before a bot sends money back, look for an open dispute on that charge. Happy to DM a free look at 90 days of refunds for any merchant you work with.

[Axiru @AxiruAI]

@RohanBhanotAI Under $25, order verified, no other refund in 30 days is a solid set of rules. Agreed, the repeat check is the hard part. It gets harder when refunds come from two places, like Shopify and a returns app. Happy to DM a free look at your last 90 days of refunds.