AI Risk Wire

Cases › operational mistake › LR-2026-0641

Who Gives A Crap suspends AI agent after email error about prices

Preliminary Why: Publisher not yet named · subject to change

Status: fix reported

Severity minor
Reported 8 Jul 2026
Industry direct-to-consumer toilet paper deliveries

No source states: loss or records, developer and model, agent tool, could do, deployer.

Severity minor: Test: contained to one deployer or one evaluation, routine cleanup, no third party touched. Fact: one customer received the wrong price email, the mistake was low-risk and quickly resolved, and no data or systems of others were stated to be touched.

Who Gives A Crap, an Australian toilet paper retailer, had an AI agent write some of its customer emails about an upcoming price rise. The agent told one customer that a bamboo toilet paper subscription would more than double in price, and a second AI-generated email reaffirmed the wrong price. The customer questioned the change with the company, which confirmed it in a reply. The company said the original email contained a typo, with the new price covering 48 rolls rather than 24. Once the problem was identified, the company shut down the email agent and sent a correcting email to the affected customer. The company said it is working on quality assurance testing for its AI agents.

What happened

1 · What it was doingThe company used an AI agent to write some customer emails.
2 · What went wrongThe agent wrote an email telling one customer their subscription price would more than double.
3 · What it reached or didThe email stated a price rise from $66 for 48 rolls to $69.50 for 24 rolls.
4 · How it was caughtThe customer questioned the price with the company.

What was done

The email agent was shut down and a correcting email was sent to the customer.
Fix reported

“Shut down the email agent and sent a correcting email to the affected customer”

Risks

Sorted by rule from the case's fields; not yet read from the sources.

Controls

Harness

Effective Stop control

Control failure chain

L1 · Model Not stated in sourcesProvider not named
L2 · Harness EffectiveHarness not named
L3 · Application Not stated in sourcesDeployer not named
L4 · Infrastructure / target Not stated in sourcesOrganisation hit

L1 · ModelProvider not namedModel provider

No source mentions a control at this layer.

Model version not linked to a model card.

L2 · HarnessHarness not namedAgent framework vendor

EffectiveStop control“Once identified, we immediately shut down the email agent and sent a correcting email to the affected customer.”SmartCompany

Other controls for this risk at this layer, not mentioned in the sources: Tool allow-list, Fail-closed defaults.

L3 · ApplicationDeployer not namedDeployer

No source mentions a control at this layer.

Other controls for this risk at this layer, not mentioned in the sources: Human approval for high-impact actions, Change review and separation of duties, Least-privilege access for the agent, Production kept apart from test, Activity logs, Adversarial testing before deployment.

L4 · Infrastructure / targetOrganisation hit

No source mentions a control at this layer.

Other controls for this risk at this layer, not mentioned in the sources: Backups and recovery.

EffectiveNot effectiveNot effective (bypassed)Did not existNot stated in sourcesNot yet checkedAttacker's ownOrder: model · harness · application · target

Controls for this risk: result in this incident

ControlPart of the systemResult
Stop controlL2 HarnessEffective
Tool allow-listL2 HarnessNot stated in sources
Fail-closed defaultsL2 HarnessNot stated in sources
Least-privilege access for the agentL3 ApplicationNot stated in sources
Production kept apart from testL3 ApplicationNot stated in sources
Human approval for high-impact actionsL3 ApplicationNot stated in sources
Change review and separation of dutiesL3 ApplicationNot stated in sources
Activity logsL3 ApplicationNot stated in sources
Adversarial testing before deploymentL3 ApplicationNot stated in sources
Backups and recoveryL4 Infrastructure / targetNot stated in sources

Evidence

No source states: when it happened, when it was found, how many records were exposed, the direct loss, a count of the other parties affected, any court case, the cause, what set it off.

Show the full account: 5 items from the primary sources

What happened, in order

  1. August 6 The customer happiness team confirmed the wrong change by email, starting from August 6.

Not dated in the sources:

  • One customer received an email wrongly stating their subscription price would more than double.
  • The customer queried the price with the company.
  • A second, AI-generated email reaffirmed the wrong price.
  • After the problem was identified, the email agent was shut down and a correcting email was sent.
Show every field, with its source and grade

The AI system

What it was for
writing some customer emails
C one secondary source states it

What happened

Earliest source we hold
2026-07-08
D inferred by us from the described behaviour, reason shown
What was stopped
email agent shut down
C one secondary source states it
What was done
Shut down the email agent and sent a correcting email to the affected customer
C one secondary source states it
What went wrong
operational mistake
C one secondary source states it
AI agent
an AI agent acted
C one secondary source states it
control chain
layer: L2; party: harness vendor; party name: Who Gives A Crap; control id: B3; outcome: held; quote: Once identified, we immediately shut down the email agent and sent a correcting email to the affected customer.; source index: 0
C one secondary source states it

Impact

Severity
minor
C one secondary source states it
Who was affected
one customer who was sent an email with a wrong subscription price
C one secondary source states it
Why this severity
Test: contained to one deployer or one evaluation, routine cleanup, no third party touched. Fact: one customer received the wrong price email, the mistake was low-risk and quickly resolved, and no data or systems of others were stated to be touched.
D inferred by us from the described behaviour, reason shown

Context

Industry
direct-to-consumer toilet paper deliveries
C one secondary source states it
Country
Australia
C one secondary source states it

Notes on gaps

The date of the article (2026-07-08) is not used as a disclosure date. The August 6 date is the effective date of the wrong change, not the incident date. Control I2 as missing is inferred from the company saying it is working on quality assurance testing. The PocketOS incident mentioned in the article is a different incident and is excluded.

LR-2026-0641 · version 1 · published 2026-11-01 after automatic checks; not yet checked field by field by a person · JSON