On a Monday morning in November 2026, a cloud operations lead opened a ticket queue already stacked with SAP NetWeaver webshell probes against internet-facing application servers. The same week, practitioner channels flagged ARTEX-style automated campaign tooling and a widespread PHP flaw expected to knock hosting fleets offline. None of those stories started as a classic perimeter breach. They landed where many teams still leave cloud tenancy soft: instance metadata open to older protocols, jump hosts with standing admin roles, and golden images that still trust whatever intermediate CA the last bake pulled in.
If your estate runs workloads on AWS, Azure, or GCP, the useful response is not another generic hardening checklist. It is a same-shift sequence that detects role abuse early, then closes the paths attackers use to turn a single compromised host into lasting control-plane reach.
What the Monday queue usually looks like
A typical first signal is not ransomware. It is repetitive HTTP requests for known SAP NetWeaver webshell paths, PHP admin panels, and leftover installer directories on cloud public IPs. Scanners move by prefix. They hit shared tenancy ranges, then pause long enough for a human operator or autopilot kit to try credential reuse against SSH, RDP, or a vendor remote-support agent.
When a probe succeeds, the next move is often local privilege escalation followed by a call to the instance metadata service. With IMDSv1 or an equivalent unauthenticated metadata path still enabled, the attacker can pull temporary credentials tied to the instance profile. Those credentials then open object storage, secrets managers, or neighboring compute APIs without touching your VPN.
Parallel noise makes triage harder. An intermediate CA compromise such as the eMudhra Intermediate CA incident forces teams to re-validate trust stores inside images and load balancers while the same operators are still chasing webshell hits. Treat those as concurrent workstreams with clear ownership, not as a reason to freeze hardening.
Detection steps that map to real cloud telemetry
1. Join edge exploit noise to compute identity
Pull WAF, reverse-proxy, and host firewall logs for the same window as SAP NetWeaver webshell path hits, PHP admin probes, and known scanner user-agents. Enrich each source IP with ASN, prefix tenure, and whether the destination host holds an instance role with write access to IAM, Key Vault, Secrets Manager, or deployment pipelines.
Flag any host that received exploit-path traffic and then showed:
- Successful local authentication from a rare source
- Outbound calls to 169.254.169.254 or the platform metadata hostname
- New API calls signed with that instance role within minutes
That three-event join is your highest-value cloud intrusion signature for this threat class.
2. Watch metadata access patterns, not only API errors
Enable and retain metadata access logs where the platform supports them. Hunt for:
- IMDSv1 (or non-token) requests from user processes outside bootstrap windows
- High-frequency credential fetches from shells, interpreters, or web workers
- Role sessions that suddenly call List*, Describe*, and GetSecret* APIs across accounts or projects they rarely touch
On AWS, require IMDSv2 and alert when PutInstanceIdentity or role assumption volume spikes after a public HTTP anomaly. On Azure, correlate IMDS token requests with unusual Managed Identity use against Key Vault and ARM. On GCP, watch metadata server hits alongside unexpected service-account key-like behavior from GCE instances that should only use short-lived tokens.
3. Inventory soft jump and break-glass paths
Enumerate bastions, vendor support appliances, CI runners, and “temporary” admin VMs that still hold broad roles. Many ARTEX-like campaigns and access brokers prefer these hosts because they already bridge management networks. Detect standing SSH keys older than your rotation policy, shared local admin passwords, and security groups that still allow 22/3389 from 0.0.0.0/0 “for emergencies.”
4. Treat image and certificate drift as a detection input
After CA compromise notices, compare the intermediate and root stores baked into AMIs, machine images, and container base layers against your current trust policy. A newly untrusted intermediate that remains in a production image is both a supply-chain defect and an attacker convenience for TLS interception or fake update endpoints. Score images that missed the last trust rebuild as higher risk when they also sit behind public listeners.
Response actions that harden the tenancy in hours
Immediate containment on a suspected role harvest
- Detach or replace the instance profile or managed identity on the affected host.
- Invalidate active role sessions and rotate any secrets the role could read in the prior 24 hours.
- Snapshot the disk for forensics, then rebuild from a known-good image rather than attempting in-place cleanup of a webshelled app tier.
- Block the source prefixes at the edge and add the role ARN or principal ID to a watchlist for lateral API use.
Keep customer-auth traffic and the remote-support plane on separate identity boundaries so a containment action on one path does not black-hole partner logins.
Same-day hardening controls
Force tokenized metadata. Require IMDSv2 hop limit 1 (or platform equivalent), deny non-token metadata from application processes, and fail pipeline checks that launch instances without the hardened setting.
Shrink instance roles to task scope. Replace account-wide read/write policies with per-service roles that can only reach the buckets, queues, and secret paths that workload needs. Ban wildcards on iam:PassRole, sts:AssumeRole, and secret list APIs for app tiers.
Close soft jump paths. Move admin access behind SSO with MFA, short-lived certificates or just-in-time elevation, and private connectivity. Remove standing public management ports. Put vendor remote support in a dedicated account or project with explicit peering and time-boxed roles.
Patch and rate-limit the public app surface. For PHP and SAP-facing fleets, apply emergency patches on a measured ring: canary hosts first, then public listeners, then internal replicas. While rings complete, apply WAF rules and ASN or prefix rate caps on exploit paths so scanners burn time without earning shells.
Rebuild trust into the image pipeline. When an intermediate CA is reported compromised, revoke it in your internal trust bundle, rebuild golden images, and redeploy internet-facing listeners that terminate TLS with the updated chain. Record image digests in the change ticket so audit can prove which tenants still run pre-rebuild builds.
Control-plane guardrails that survive the next scout wave
- Organization-level policies that deny public management ports and require encrypted disks and tokenized metadata on new compute
- CloudTrail, Activity Log, or Admin Activity sinks with immutable retention and alerts on Disable*, DeleteTrail, and unusual AssumeRole chains
- Separate CI deploy roles from runtime roles so a compromised web worker cannot push a new stack
- Network policies that keep metadata-equivalent endpoints unreachable from containers that only need app traffic
A practical tabletop for the next advisory week
Run a two-hour exercise with cloud ops, identity, and app owners. Inject three facts: SAP webshell path hits against a public app subnet, a PHP critical advisory timed to the same day, and a peer note that an intermediate CA in your image trust store is under review. Require the team to produce, before the exercise ends:
- The detection query that joins exploit paths to metadata and role API use
- The containment runbook that rotates the role without taking down partner auth
- The change list that forces IMDSv2/equivalent, shrinks two overprivileged roles, and rebuilds one public image with an updated trust store
Score the exercise on time-to-role-detach and on whether public listeners still accepted management protocols from the open internet at the end of the window.
Takeaways to put on the shift handoff
Webshell scouts and autopilot kits succeed in cloud estates when metadata, jump hosts, and image trust lag behind the edge. Detect by joining public exploit noise to metadata fetches and sudden role API use. Respond by detaching the role, rebuilding the host, forcing tokenized metadata, shrinking identities, and closing standing admin exposure. Keep certificate and PHP emergency work on parallel tracks with named owners so a CA notice or mass outage advisory does not stall role containment.
Print the three-event join (exploit path, metadata access, anomalous role API) next to your on-call board. That single hunt turns a noisy Monday of SAP, PHP, and campaign chatter into a concrete cloud hardening drill you can finish before the next scout wave arrives.