LLMjacking: Customer report

Nicholas (Niccolo) Roberts · Sovereign Source Super Intelligence

Expanded web report · Related PDF snapshot: 22 October 2026 · PDF page 11

Research reviewed 22 October 2026 · Evidence and editorial planning guidance are distinguished below.

Understand your account and use a verified support route

Use authenticated, official AI features and keep private information out of public or unfamiliar endpoints. If a service behaves unexpectedly, stop and report the problem rather than trying repeated prompts that could worsen the situation.

Your immediate priorities

Choose the business’s human support option when AI is unavailable.
Report unexplained account usage using the provider’s established process.

How to work through the decision

A public incident report cannot determine what happened to your account. Ask the business to explain confirmed effects, remaining uncertainty and any action relevant to the service you actually use. Keep useful order references or notices and contact support through a known channel. Treat urgent unsolicited requests for credentials or payment as unverified until you confirm them independently.

Evidence to ask for

A clear service alternative and account-specific guidance where needed.

Is this an approved service with a responsible operator and a clear privacy explanation?

Work with the next person

Give customer service the relevant order or product reference and a concise description of the issue. Do not send passwords, one-time codes or full payment details. Ask for an accountable follow-up if the answer is still under investigation.

Continue with the customer service report for this topic.

Before following an instruction

Confirm that the requested action comes from the organization you use and applies to your account or product version. If the instruction is unclear, ask for an explanation through the verified support route. A public planning article is not an account-specific notice.

What a useful outcome looks like

A clear service alternative and account-specific guidance where needed. Keep the decision dated and return to it if new evidence changes the scope.

Sources and further reading

Symantec/Broadcom LLMjacking research

What does this mean for your business?

Use the Cyber Health Review to turn the research into priorities for your own operating model.

Contact help@cyberpolycrisis.com

A planning tool.
A starting point for evidence.

Based on the Cyberpolycrisis articles by Nicholas Roberts. The articles examine shared dependencies across concurrent events; they do not establish a single coordinated campaign.

How is readiness calculated?

Each applicable control has equal weight: established and tested = 100, partly in place = 50, missing or untested = 0, and not sure = 0 pending verification. The mean is rounded to a whole number. AI controls are excluded when you report no AI use.

The planning bands are 0–49 “Stabilize”, 50–79 “Close gaps”, and 80–100 “Close remaining gaps” or “Validate and maintain” when none remain. These are editorial priorities, not validated risk thresholds. A current incident concern overrides the band. A high average can still contain a consequential individual gap.

This is a self-reported checklist. It does not measure the probability of a breach, scan your systems or certify your security.

What does the network map show?

Connections reflect thematic overlap in the published network analysis. Highlighted connections mean your answers show gaps in both areas. They do not demonstrate a live attack path, causal link or measured likelihood. Line thickness is illustrative.

Why these platforms?

The shortlist matches your business needs to documented capabilities. Shopify can own commerce records; Palantir can support governed operational workflows; Databricks can govern analytics and AI data; Cloudflare can protect access and applications and host suitable workloads.

“Safe haven” means a better-controlled operating model, not a guarantee from a vendor. Documentation alone does not establish a comparative security track record. Before buying, review current assurance reports, relevant incident disclosures, contract terms, data location, implementation cost and exit options.

Vendor controls depend on product, plan and configuration. Your team remains responsible for access, integrations, data and recovery.

Where do my answers go?

Scoring runs in your browser. Saving progress is optional and uses this browser’s local storage. Anyone with access to the browser profile could read saved answers. Your completed review is sent to the chatbot only when you explicitly select “Include my completed review with the next question.”

Chat sends your message and recent conversation to Cloudflare Workers AI. This app keeps the conversation in page memory and does not save transcripts to a database. “New chat” clears that local conversation; it cannot recall data already sent. Hosting and AI providers process requests under their own terms. Do not share passwords, API keys, customer data or confidential incident details.

If you choose “Email me a follow-up,” we store your email address, consent version and request timestamps privately in Cloudflare for Nicholas Roberts to respond. Your chat and assessment aren’t included in this request, and the email address isn’t sent to the AI. This is a personal follow-up request, not a newsletter subscription. You can ask Nicholas to remove your request through his LinkedIn profile.

Downloads stay on your device. The contact button opens a generic email draft and does not include your answers. Hosting providers still process ordinary web requests; do not put assessment information in URLs.

Sources and version

Methodology 1.0. Capability documentation reviewed October 2026. Review current vendor terms and controls before making a decision.

JavaScript is needed to calculate your plan locally. You can still download the free report, read the methodology and explore the Cyberpolycrisis network analysis.