AI is 'both the weapon and the target' in latest wave of cyberattacks

1. Executive summary

CrowdStrike's latest Threat Hunting Report documents a 89% rise in AI-enabled adversary activity during 2025, with AI infrastructure now serving as both an attack tool and a high-value target. Two campaign patterns are directly relevant to EMEA financial services: North Korean actor Famous Chollima (attribution unconfirmed — no MITRE ATT&CK profile) deployed trojanised GitHub repositories against cryptocurrency and blockchain firms, executing hidden scripts on developer clone to gain environment access; and financially motivated crew Altered Spider (attribution unconfirmed — no MITRE ATT&CK profile) compromised 300+ software dependencies in a single day, harvesting credentials and pivoting to cloud environments for theft and extortion. The report also documents LLMjacking — credential theft to access frontier-model APIs — with one token thief sending ~200,000 API requests in two minutes. CrowdStrike assesses that AI-assisted exploitation now occurs within 24–48 hours of PoC publication, rendering traditional 30-day patch windows obsolete.

2. Regulatory framing

No specific DORA/NIS2 article is directly engaged by this item. The advisory describes emerging threat trends and campaign patterns rather than a specific incident at a regulated entity. The general duty to maintain ICT-related incident management processes (DORA Art. 17) and conduct operational resilience testing (DORA Art. 24) applies contextually, but no distinctive trigger fact in this item changes what a client must do under a specific article beyond baseline obligations that apply to any threat intelligence.

3. Technical analysis & attack chain

Campaign 1 — Famous Chollima supply-chain compromise (January–February 2026)

1. Initial access via trojanised repositories. Famous Chollima published malicious repositories on GitHub containing legitimate-looking project files alongside hidden malicious scripts. Targeting was directed at cryptocurrency and blockchain companies, specifically AI-focused development environments.
2. Execution on clone. When developers opened/cloned these repos, malicious scripts automatically executed commands, granting the actor access to their development environments.
3. Insider threat enablement. CrowdStrike reports this crew created "entire fake companies with AI-generated websites, GitHub accounts, and email infrastructure to support insider threat operations" — extending the known fake-IT-worker tradecraft with AI-generated supporting infrastructure.

CrowdStrike assesses Famous Chollima as the North Korean group demonstrating "the most advanced AI usage" across H2 2025–H1 2026. The group is described as a sub-unit operating under the Lazarus Group (MITRE G0032) umbrella. Attribution caveat: Famous Chollima has no MITRE ATT&CK profile; treat the attribution as unconfirmed. A separate Lazarus offshoot, tracked as Stardust Chollima / Sapphire Sleet, is suspected behind the March Axios supply-chain attack — this is also single-sourced via CrowdStrike and should be treated as unconfirmed.

Campaign 2 — Altered Spider dependency compromise

1. Mass software dependency compromise. Altered Spider (also tracked as TeamPCP) compromised more than 300 software dependencies in a single day, targeting developers' AI tools.
2. Credential and secret harvesting. The actor harvested credentials and secrets from compromised development environments.
3. Cloud pivot. Pivoted from endpoint to cloud environments within minutes — Meyers stated the actor "hits the endpoint in seconds and within minutes, they're inside of the cloud."
4. Theft and extortion. Final objectives included data theft and extortion.

Attribution caveat: Neither Altered Spider nor TeamPCP have MITRE ATT&CK profiles; attribution is unconfirmed.

Campaign 3 — LLMjacking / cost harvesting

1. Credential theft. Criminals steal corporate credentials with access to frontier-model APIs.
2. API abuse. Stolen credentials are used to run unauthorised API requests against LLM services. CrowdStrike documented one token thief sending ~200,000 API requests in two minutes.
3. Cost harvesting. Deliberate inflation of victim AI usage to run up billing costs.

AI-accelerated vulnerability exploitation

CrowdStrike reports that from January to June 2026, 88% of observed exploitation using public PoC code occurred within 48 hours of PoC release. China-linked groups Vault Panda and Genesis Panda (both unconfirmed — no MITRE ATT&CK profiles) launched attacks within 24 hours of disclosure. CrowdStrike attributes this acceleration to AI-assisted weaponisation of vulnerabilities.

CVE volume context: ~48,200 CVEs registered in 2025; ~43,000 already recorded by late July 2026; June 2026 alone saw 7,600+ CVEs.

4. Mitigation & containment

P1 — Within 24 hours

• GitHub repository hygiene: Audit all developer workstations and CI/CD pipelines for repositories cloned from unverified or newly-created GitHub accounts. Block execution of post-clone scripts (e.g., disable Git auto-execution hooks, enforce git config --global init.templateDir to a controlled template that strips hook execution). Review any repositories associated with AI/blockchain/cryptocurrency development for unauthorised scripts.
• Secret scanning: Enable secret scanning across all repositories and CI/CD pipelines. Rotate any credentials, API keys, or tokens stored in repositories that may have been exposed to trojanised dependencies. Prioritise cloud provider credentials (AWS, Azure, GCP) and LLM API keys.
• LLM API access controls: Review API key usage logs for anomalous request volumes. Implement rate-limiting on frontier-model API keys. Alert on any key exceeding 10,000 requests per hour. Revoke and rotate keys with no documented owner or usage pattern.

P2 — Within 72 hours

• Dependency integrity review: Conduct a retrospective audit of npm packages and GitHub dependencies introduced in the past 18 months. Amazon attributed four npm compromises to the same North Korean crew over this period — assess whether any affected packages are present in your environment.
• Cloud access review: Review IAM roles and service accounts for credentials that may have been harvested via compromised development tools. Enforce MFA on all cloud console and API access. Implement conditional access policies restricting API calls to known IP ranges and device postures.
• Developer onboarding controls: For organisations employing or contracting remote developers, enhance background verification. Famous Chollima's fake-IT-worker operations now use AI-generated company websites, GitHub accounts, and email infrastructure — standard employment checks may not detect synthetic identities. Cross-reference GitHub commit history, account age, and contribution patterns for anomalies.

P3 — Within 7 days

• Patch cadence review: Given the 24–48 hour exploitation window, reassess vulnerability management SLAs. Prioritise internet-facing services for 48-hour patching from PoC publication, not 30-day cycles. Establish automated monitoring for PoC releases affecting your exposed technology stack.
• AI infrastructure hardening: Treat AI/ML pipelines as critical attack surface. Restrict CI/CD pipeline permissions. Implement signed dependency verification (e.g., npm provenance, Sigstore/cosign for container images). Network-segment development environments from production cloud infrastructure to slow cloud pivot speed.
• Threat hunting: Hunt for LLMjacking indicators — review cloud billing for unexpected spikes in AI service usage; correlate API gateway logs against known user baselines; investigate any service account making high-volume LLM API calls outside business hours.

5. Indicators of compromise

No indicators of compromise available in the source material.

Behavioural indicators

6. Detection

Insufficient indicators to author detection rules. The source material describes campaign behaviours and actor tradecraft but does not provide specific file hashes, filenames, registry keys, mutex names, command-line strings, or network indicators suitable for YARA or Sigma rule construction.

Threat actor context

Lazarus Group · G0032 · aka Labyrinth Chollima, HIDDEN COBRA, Guardians of Peace, ZINC, NICKEL ACADEMY

Lazarus Group is a North Korean state-sponsored cyber threat group attributed to the Reconnaissance General Bureau (RGB). Lazarus Group has been active since at least 2009 and is reportedly responsible for the November 2014 destructive wiper attack on Sony Pictures Entertainment, identified by Novetta as part of Operation Blockbuster. …

APT38 · G0082 · aka NICKEL GLADSTONE, BeagleBoyz, Bluenoroff, Stardust Chollima, Sapphire Sleet

APT38 is a North Korean state-sponsored threat group that specializes in financial cyber operations; it has been attributed to the Reconnaissance General Bureau. Active since at least 2014, APT38 has targeted banks, financial institutions, casinos, cryptocurrency exchanges, SWIFT system endpoints, and ATMs in at least 38 countries worldwide. …

No MITRE ATT&CK profile for: Famous Chollima, Altered Spider, TeamPCP, Vault Panda, Genesis Panda.

7. Sources

• The Register, "AI is 'both the weapon and the target' in latest wave of cyberattacks," https://www.theregister.com/cyber-crime/2026/08/03/ai-is-both-the-weapon-and-the-target-in-latest-wave-of-cyberattacks/5281534, 2026-08-03

8. Adverse Trace position

This advisory is based on a single source (The Register reporting on CrowdStrike's Threat Hunting Report); the underlying technical detail has not been independently corroborated. The threat patterns described — AI-enabled supply-chain compromise, LLMjacking, and compressed exploitation timelines — are consistent with observed industry trends and warrant proactive defensive action. Attribution to Famous Chollima, Altered Spider/TeamPCP, Vault Panda, and Genesis Panda is unconfirmed (no MITRE ATT&CK profiles for these group names); the Lazarus Group parent attribution (G0032) is established. EMEA financial services clients with AI/ML development pipelines, cloud-hosted LLM API access, or cryptocurrency/blockchain operations should treat this as a high-priority awareness and control-gap assessment item. We will update this advisory if CrowdStrike publishes the full report with atomic IOCs or if independent corroboration emerges.

Read the original source →

Published via PulseTrace — Adverse Trace threat intelligence.