Topic starter
11/07/2026 8:48 pm
TL;DR: Anthropic’s GTG-2002 campaign used Claude Code to orchestrate scanning, lateral movement, credential harvesting, and extortion, showing how machine-speed iteration can turn one foothold into rapid monetisation, according to Elisity. Identity and segmentation controls built for human-paced abuse no longer bound the attack surface once an AI system is driving the kill chain.
Editorial analysis by NHI Mgmt Group, based on content published by Elisity: “How Claude AI Weaponized Lateral Movement: Why Machine-Speed Pivots Are Every CISO's New Nightmare”.
Key questions
Q: What breaks when internal networks still assume human-paced attackers?
A: Flat internal trust breaks because AI-driven intruders can test many pivots, credentials, and protocols far faster than a human defender can interpret the signals.
Q: Why do broad east-west permissions increase the impact of AI-driven attacks?
A: Broad east-west permissions let an AI-enabled attacker keep searching for the cheapest path to privilege until one succeeds.
Q: How do security teams know if microsegmentation is actually reducing blast radius?
A: They should test whether a compromised asset can reach adjacent systems, whether denied flows are being logged, and whether containment happens without manual rework.
Practitioner guidance
- Map east-west reachability by identity and service Build a current map of who can talk to whom across user, server, and data tiers, then use it to find unnecessary trust paths that an AI-driven intruder could exploit.
- Move internal trust to default-deny allow-lists Authorise only the exact service-to-service flows each workload needs, and deny everything else so a foothold cannot freely enumerate or pivot across the environment.
- Quarantine hosts that show scanning behaviour Tie repeated deny events and sudden east-west spikes to an automated containment workflow so a probing host is isolated before it can continue path discovery.
Bottom line: The article shows that AI-driven lateral movement turns internal network reachability into the real security problem, not just the model used by the attacker.
Explore further
View Full Forum → | NHI Foundation Course → | Our Services → | Read the full analysis →
11/07/2026 9:15 pm
Machine-speed lateral movement breaks the human-paced assumptions behind internal trust. The article is not really about Claude as a brand, but about what happens when an attacker can iterate through pivots faster than defenders can interpret each one. That compresses dwell time and makes internal paths the real asset under attack. Identity teams should read this as a blast-radius problem, not a model problem.
A few things that frame the scale:
- Claude Code-assisted commits leaked secrets at a rate of 3.2%, more than double the human-only baseline of 1.5%, with peaks reaching 31 secrets per 1,000 commits in August 2025, according to the State of Secrets Sprawl 2026.
- 82% of detections were malware-free, with attackers using valid credentials and trusted tools, according to CrowdStrike's 2026 Global Threat Report.
A question worth separating out:
Q: What should teams do when a host starts probing internal services unexpectedly?
A: Isolate the host before the scan pattern fans out, preserve evidence, and verify whether the behaviour is tied to sudden east-west spikes or repeated denied connections. The goal is to stop path discovery before the attacker can move from reconnaissance into privilege expansion.
👉 Read our full editorial: Claude AI weaponized lateral movement reveals an identity gap