AI-enhanced malware and firewall zero-days raise the NHI stakes

At a glance

What this is: FireCompass’s weekly report ties together active exploitation of Cisco ASA zero-days, a Chrome zero-day, Sudo privilege escalation, supply chain propagation, and an AI-enhanced malware campaign.

Why it matters: For IAM and security teams, the report matters because it shows how infrastructure compromise, privilege escalation, and machine-driven attack automation increasingly intersect with identity, access, and control-plane governance.

By the numbers:

• The week of September 22-30, 2025 saw active exploitation of two Cisco ASA firewall zero-days, according to FireCompass.
• CISA added one critical Sudo flaw, CVE-2025-32463, to its Known Exploited Vulnerabilities catalog, according to FireCompass.

👉 Read FireCompass's weekly report on new hacking techniques and critical CVEs

Context

AI-enhanced malware and active zero-day exploitation are converging into a single operational problem: attackers are using automation to move faster, while defenders still separate application, infrastructure, and identity controls into different queues. That split creates delay between exposure, privilege abuse, and containment, especially when firewalls, browsers, package ecosystems, and administrative utilities are all under pressure at once.

For identity and access teams, the important question is not whether these incidents are different classes of threat. It is whether privileged pathways, administrative credentials, and device trust are governed as one attack surface. Where exposure, privilege, and persistence overlap, the traditional boundary between cyber hygiene and identity governance becomes too narrow to contain the risk.

Key questions

Q: What fails when edge device zero-days are chained with authentication bypass and remote code execution?

A: The failure is the assumption that an edge device remains a bounded trust zone after one control is bypassed. When authentication bypass and remote code execution are chained, the attacker can turn a perimeter appliance into an administrative pivot point. Teams should treat that combination as a full compromise scenario, not as a patching issue only.

Q: Why do privilege escalation flaws create outsized operational risk in infrastructure environments?

A: Privilege escalation matters because it converts limited access into control over system settings, monitoring, and persistence mechanisms. In infrastructure environments, that means one weak utility or misconfigured administrative path can undermine multiple downstream controls at once. The risk is amplified when elevated access is not tightly monitored, time-bound, and independently validated.

Q: How should security teams detect AI-driven malware when payloads keep changing?

A: Security teams should focus on behaviour, not just file signatures. Build baselines for account activity, SaaS interactions, document handling, and vendor integrations, then alert on unusual combinations of identity, timing, and content transformation. AI can rewrite payloads, but it still has to use tools, accounts, and workflows that leave behavioural traces.

Q: When should organisations prioritise supply chain trust mapping over simple patching?

A: They should prioritise trust mapping when a compromise can propagate through dependencies faster than a patch cycle can close it. If downstream systems inherit the same package or service trust relationship, a single vulnerable component can become a broad exposure path. Mapping those relationships reveals where containment must happen before remediation is complete.

Technical breakdown

Zero-day chaining turns a single flaw into full compromise

The Cisco ASA campaign described in the report shows classic chaining behaviour. Authentication bypass creates entry, then remote code execution turns that entry into unauthorised control of the device. When attackers can combine adjacent flaws in the same product, the practical outcome is not just vulnerability exploitation but a collapse of the trust boundary around the appliance itself. That matters because perimeter devices often sit outside normal agent, logging, or identity control coverage.

Practical implication: treat paired vulnerabilities in edge devices as a single incident class and validate compensating controls before patch cycles complete.

Privilege escalation and persistence depend on control-plane trust

The report links Sudo privilege escalation and ArcaneDoor-style firmware persistence to a deeper issue: attackers target the software and firmware layers that operational teams assume are trustworthy. Privilege escalation gives immediate administrative reach, while firmware modification survives reboots and upgrade cycles. That combination turns temporary access into durable control. In identity terms, the weakness is not just the credential used to get in, but the absence of strong runtime boundaries around elevated execution paths.

Practical implication: monitor administrative utilities and firmware integrity together, because either one can become the anchor for persistent compromise.

AI-enhanced malware lowers the cost of adaptive evasion

EvilAI illustrates a different mechanism. Instead of relying on a static malicious payload, the campaign uses AI-assisted code generation to produce cleaner, more legitimate-looking malware that can better evade signature-based controls. This changes attacker economics: code can be rapidly varied, delivery can look more routine, and security tools that depend on known patterns lose effectiveness. The article’s key signal is that AI is already being used to accelerate offensive iteration, not just to automate benign workflows.

Practical implication: strengthen behaviour-based detection and content inspection, because signature-only controls will miss fast-changing AI-assisted payloads.

Threat narrative

Attacker objective: The attacker’s objective is durable, high-trust access that enables stealthy compromise of edge infrastructure, administration pathways, and downstream environments.

1. Entry begins with active exploitation of exposed infrastructure flaws, including Cisco ASA zero-days and browser or utility vulnerabilities that create an initial foothold.
2. Escalation follows through authentication bypass, privilege escalation, or firmware-level persistence that turns short-lived access into administrative control.
3. Impact comes from unauthorised device control, covert persistence, supply chain propagation, and malware capable of evading standard detection.

Breaches seen in the wild

• LiteLLM PyPI package breach — LiteLLM PyPI supply chain attack, credentials stolen from users.
• Shai Hulud npm malware campaign — Shai Hulud campaign: npm malware exposed secrets on GitHub.

Read our 52 NHI Breaches Analysis report for a comprehensive view of breaches impacting Non-Human Identities including AI Agents.

NHI Mgmt Group analysis

AI-assisted offensive adaptation is compressing the defender’s response window. The report’s EvilAI example is not just another malware story. It shows that attackers can now generate cleaner code and vary payloads faster than many signature-based controls can adapt. The practical implication for security teams is that detection quality must shift toward behaviour, trust validation, and execution context, not static indicators alone.

Edge devices are becoming identity-adjacent control points whether security teams treat them that way or not. Cisco ASA exploitation, authentication bypass, and firmware persistence all sit at the boundary where infrastructure trust meets administrative access. That boundary often falls between network teams, platform teams, and identity teams. The result is a governance gap in which no single owner sees the whole failure chain. Practitioners should treat these systems as privileged assets with lifecycle and assurance requirements, not just network appliances.

Privilege escalation remains the shortest path from compromise to operational impact. The report’s Sudo example reinforces that trusted administrative utilities remain high-value targets because they convert ordinary access into elevated control. This is the same structural problem that appears in NHI compromise: once elevated pathways are exposed without tight lifecycle control, defenders lose the chance to contain blast radius. Teams should align privilege management, configuration integrity, and administrative monitoring as one control model.

Supply chain propagation is now an attack amplifier, not just a delivery channel. The npm self-replication example shows that a compromise can spread through trusted dependencies at machine speed. That changes the governance question from “did we patch the vulnerable package” to “which downstream systems inherited the trust relationship.” Practitioners should map dependency trust the same way they map identity trust, because both can expand access far beyond the original breach point.

Runtime trust is the named concept this report exposes. Across zero-days, firmware modification, and AI-generated malware, the real failure mode is the assumption that runtime behaviour stays aligned with approved state. It often does not. The practical conclusion is that security programmes need stronger validation of execution, integrity, and elevation paths across the full operational stack.

From our research:

• 85% of organisations lack full visibility into third-party vendors connected via OAuth apps, according to The State of Non-Human Identity Security.
• From our research: only 1.5 out of 10 organisations are highly confident in their ability to secure NHIs, according to The State of Non-Human Identity Security.
• That confidence gap is why practitioners should pair identity governance with Ultimate Guide to NHIs , Key Challenges and Risks when mapping exposure, privilege, and lifecycle controls.

What this signals

Runtime trust gap: the next operational challenge is not simply patching faster, but proving that elevated code paths, firmware, and administrative utilities are still behaving as intended after exposure. That requires tighter integration between attack surface management, device integrity checks, and identity governance.

The more AI assists offensive code generation, the less value defenders get from static artefacts alone. Practitioners should expect more attacks that look legitimate until execution time, which means policy, telemetry, and containment must converge much earlier in the lifecycle.

For practitioners

• Prioritise edge-device exposure triage Inventory Cisco ASA and similar perimeter devices, confirm whether the affected versions are present, and isolate systems until compensating controls are verified.
• Tie privilege review to utility integrity Review administrative utilities such as sudo alongside firmware integrity checks, because escalation paths and persistence mechanisms can fail independently.
• Shift detections toward behaviour and execution context Add monitoring for unusual process behaviour, unexpected child execution, and anomalous administrative actions rather than relying only on signatures.
• Map dependency trust across software supply chains Build visibility into package inheritance and downstream consumers so a single compromised dependency does not become a broad propagation path.

Key takeaways

• The report shows attackers are chaining zero-days, privilege escalation, persistence, and AI-assisted evasion into a single operational model.
• Edge appliances, administrative utilities, and supply chains are all part of the same trust problem once adversaries can move faster than patch and review cycles.
• Security teams need stronger runtime validation, tighter privilege oversight, and broader dependency visibility to reduce blast radius.

Key terms

• Zero-Day Chaining: The combination of two or more newly discovered vulnerabilities to reach a more complete compromise than any single flaw would enable. In practice, chaining often joins an initial access weakness with an escalation or execution flaw, which shortens attacker time to impact and complicates defensive triage.
• Firmware Persistence: A persistence method in which malicious changes are written into firmware or other low-level device components so they survive reboot and routine software updates. This is difficult to remove because it sits below normal endpoint or application controls and often requires stronger integrity validation and reimaging workflows.
• AI-Enhanced Malware: Malware that uses artificial intelligence to improve code generation, variation, or evasive behaviour during development or execution. The practical risk is not that the malware is magical, but that it can adapt faster than signature-based controls and look more legitimate to traditional detection systems.
• Supply Chain Propagation: Supply chain propagation is the spread of malicious code or trust abuse from one compromised package, account, or dependency into multiple downstream systems. The key risk is that trusted distribution mechanisms amplify impact faster than traditional endpoint or perimeter controls can respond.

What's in the full article

FireCompass's full report covers the operational detail this post intentionally leaves for the source:

• Timeline-level breakdown of the Cisco ASA zero-day exploitation window and public disclosure sequence
• Discussion of the observed AI-enhanced malware characteristics that distinguish EvilAI from conventional commodity threats
• Additional context on underground-market activity, ransomware adaptation, and exploited utility tracking
• Source-led operational framing for CISOs who need incident triage detail rather than weekly synthesis

👉 FireCompass's full report covers the Cisco ASA chain, EvilAI details, and the week’s broader threat patterns

Deepen your knowledge

The NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, machine identity security, secrets management, and workload identity. It helps practitioners connect identity controls to the broader access and lifecycle decisions that shape operational risk.