@hulumi/baseline CVE-2026-48035

Requires that the target AWS account was provisioned with @hulumi/baseline AccountFoundation prior to 1.4.0, AND that the attacker holds an IAM identity in that account with s3:DeleteObject or s3:DeleteObjectVersion on the audit-delivery bucket (or the ability to invoke pulumi destroy on a stack where the consumer set logBucketForceDestroy:true). For sandbox-tier deployments no additional configuration is needed because Object Lock, access logging, and the CloudTrail-Lake EventDataStore are all absent by default. Exploitation is LIMITED by: (1) requiring existing access to the AWS account (not remote/unauthenticated - there is no internet-facing component), (2) only the startup-hardened and sandbox tiers being in scope, (3) authentication requirements not confirmed from available data because no CVSS vector is published, and (4) the attack producing forensic erasure rather than direct asset compromise - value depends on the attacker already having performed actions worth concealing.

No CVSS vector, EPSS score, or KEV listing is provided, so quantitative severity signals are absent; the vendor advisory rates the issue High. This is a forensic integrity / log tamper-resistance weakness rather than an initial-access vulnerability - exploitation does not yield code execution or data exfiltration directly, but it enables an attacker who has already obtained s3:DeleteObject* (or who can trigger pulumi destroy with logBucketForceDestroy=true) to erase the CloudTrail and Config evidence trail of their activity. Real-world risk is therefore highly contextual: organizations relying on AccountFoundation's startup-hardened tier for compliance-grade audit immutability (SOC 2, PCI, breach-investigation defensibility) face material risk, while sandbox-tier deployments are essentially unmonitored from an immutability standpoint. No public exploit identified at time of analysis. Note the CWE-1059 classification (documentation/behavioral inconsistency) is unusual for a security advisory and reflects that the defect is a mismatch between promised and actual behavior - defenders should treat it as a misconfiguration-class issue with audit-integrity consequences.

An insider or a compromised IAM principal that holds s3:DeleteObject* on the AccountFoundation audit bucket pivots after performing malicious activity, deletes the CloudTrail and AWS Config objects covering the activity window, and leaves no forensic trail because no Object Lock prevents the deletion. Alternatively, an operator with privileges to run pulumi destroy against a downstream stack that sets logBucketForceDestroy:true purges every audit-log object during a routine teardown, either accidentally or as a deliberate cover-up. No public exploit identified at time of analysis; exploitation is operational rather than technical and requires no specialized tooling.