Make trusted routes easy
Give employees a known route to approved tools and a simple way to question unexpected instructions.
What this does—and does not—establish
Approval of a platform does not validate each creator's content.
AI-themed lureCASE STUDY / Reported by Island
Fake ChatGPT help led toward malware
Island observed deceptive Custom GPT replies directing people toward a Windows malware lure.
Source: IslandObservation dates, not confirmed campaign boundaries.
THE 10-SECOND TAKEAWAY
A trusted address does not make every instruction trustworthy.
HOW IT WORKED
Follow the chain.
Search
→
Advertisement
→
ChatGPT page
SIMPLIFIED VIEW · 1 / 3Reply
→
Backup site
→
False verification
SIMPLIFIED VIEW · 2 / 3Instruction
→
Human action required
→
Possible infection
SIMPLIFIED VIEW · 3 / 3STEP 03
Harmful instructions
The lure asked users to run commands; researchers analyzed the payload separately.
IslandMove through the story at your own pace.
1 / 3
THINK OF IT THIS WAY
Editorial analogy: a stranger pins a convincing notice inside a familiar building. The building's name does not authenticate the notice.
An analogy for the mechanism; not an additional claim about the incident.KEEP THE EVIDENCE IN VIEW
What we know.
What we don’t.
Reported outcomes
Paid-search links reached lookalike GPTs; repeated replies promoted an external backup site. Island
Important limits
Infections and losses are unconfirmed.
Researchers inspected responses, not GPT configurations.
No ChatGPT or Google vulnerability was established.
The lesson beyond
this one case.
Editorial lesson: check who created advice and what it asks you to do. A familiar platform can host an unfamiliar person's instructions.
Where were the people?
People created the lure and needed users to follow instructions. Island
What happened in response? +
No campaign-wide removal outcome was reported. Island
What could help
an organization?
Editorial connections for organizations protecting their own users and devices; no control guarantees prevention.
Connect the evidence
Investigate suspicious conversation output together with browser navigation and device alerts, within authorized monitoring.
What this does—and does not—establish
A visited page alone does not prove that harmful software ran.
Editorial connections to relevant controls, not evidence that a particular technology would have prevented this case. Each guide links to the security guidance behind its recommendations.
Explore the full AI security frameworkTRACE IT TO THE SOURCE
Read the evidence.
Explore more original accounts in the source report library ↗.
These are source-reported findings. An independent assessment, when available, is labeled explicitly.
01How Attackers Use Sponsored Search & Custom GPT’s in Malware DeliveryIsland · Oct 1, 2026 · security researcher investigation
Read this case’s update history →
Reviewed Oct 1, 2026 · Editorial methodology · Structured data