How we use AI

The passive reconnaissance panel on our homepage shows what any web server learns about a visitor from the connection alone. When you ask it to analyse your exposure, the narrative is written by Claude at the network edge. The model receives only the connection metadata shown in the panel, nothing is stored, and every analysis is labelled as AI-generated. If Claude cannot be reached an open model takes over and the analysis is still labelled as AI-generated.

Strikeguard, our graded security posture report for verified domains, runs entirely at the edge: the checks are deterministic and read only public configuration, and an AI model writes the executive summary and the remediation plan from those results.

Our engineers work with Claude Code day to day: building and reviewing the code behind this site and our internal tooling, reviewing our own code for security defects, and drafting documentation. An engineer reviews and owns every change before it ships.

We are building Claude-powered assistants for our analysts. These are in development and are introduced to client work only under the client’s written agreement:

• Incident triage. Turning raw alerts, logs and responder notes into a structured timeline, a containment checklist and a status update for executives, for the on-duty analyst to review and send.
• Evidence correlation. Reasoning across endpoint, identity and cloud telemetry to propose hypotheses and the next evidence to pull.
• Findings and reporting. Verifying, de-duplicating and ranking penetration-test findings by root cause and impact, and drafting remediation guidance a tester finalises.
• Incident intake. A connector that lets teams already working in Claude open an incident with us directly.

• A named person is accountable. AI drafts; a qualified analyst reviews, changes and signs off on every recommendation or deliverable that reaches a client.
• AI output is disclosed. Anything on this site that a model wrote is labelled as such, and clients are told when AI contributed to advice they receive.
• Client data stays controlled. No client data is processed by a model without written agreement covering scope, retention and the provider involved.
• Authorised use only. Offensive security work is performed solely under written authorisation and within an agreed scope, and we use AI providers in line with their usage policies, including Anthropic’s. See our security and disclosure policy.
• Models we use. Claude by Anthropic for reasoning and drafting; open models at the edge as a fallback.

Ask us anything about how AI is used in an engagement at info@cyberlinksec.com.