Tamper Diff: flag agent code changes that weaken logging, alerts or monitoring

Nobody I found sells the specific thing: a diff classifier that recognises when a PR weakens the observability and control surface. That surface spans app-level log calls and log levels, audit-table writes, alert thresholds and routing, sampling and retention settings, monitor definitions, and feature flags that switch off fraud rules. The tool would route those PRs to a second approver and keep a record that holds up in an audit. The nearby players each cover a piece. Apiiro does material change detection, but around auth, APIs and PII. MergeShield, Skylos and Endor gate agent PRs, but by general risk. IaC scanners such as Datadog Code Security and KICS check single settings like CloudTrail enable_logging. Datadog SIEM catches some silencing of alerts at runtime, but only inside Datadog. The weekly monitoring drift report looks even emptier. Today you would assemble it yourself from terraform plan output and vendor audit trails, and nothing I found normalises posture across Datadog, PagerDuty, Splunk and cloud log retention into one report a CISO can read. A new entrant could win with a fintech focus. That means mapping evidence to SOC 2, PCI DSS 10 and SOX change-management controls, plus knowing fintech details like fraud-rule flags and ledger audit tables, and selling as a cheap GitHub app with no ASPM rollout. The main risk is that this becomes a checkbox feature. Apiiro, Endor, CodeRabbit or GitHub could add an observability-weakening rule pack fairly easily. So the lasting value probably sits in the cross-tool posture baseline and the audit evidence trail, not the PR classifier on its own.

• Apiiro (Material Code Change Detection)

  The closest incumbent. Apiiro classifies PRs as material when they touch security controls, auth, data models or APIs, and can block the PR or open a ticket. Its marketing is about authentication and PII exposure. I found nothing describing rules for logging, alerting or monitoring changes, and it is an expensive enterprise ASPM product.

• MergeShield

  Risk-scores PRs from AI coding agents on GitHub across six dimensions, tracks how much each agent can be trusted, auto-merges low-risk PRs and holds high-risk ones for human review. It covers the agent-PR gating workflow in general. Its listing says nothing about observability or audit-log changes specifically.

• Skylos (open source)

  An open-source tool that added a required check flagging risky agent changes such as weakened tests, removed auth and changed guardrail settings. The check only passes after a human with write access approves the head commit. This is the same second-approver pattern, but it does not target logging or monitoring.

• Endor Labs AI Code Governance

  Reviews every PR for security and puts policy and an audit trail on every coding agent. It is a well-funded AppSec vendor that could add control-surface rules.

• Datadog Code Security IaC rules / KICS

  Static IaC rules that flag Terraform setting CloudTrail enable_logging to false, and KICS has the same check under an Observability category. They cover a narrow slice of infra logging config. They look at a single state, not at whether a diff weakened something, and they ignore application-level logging, alert thresholds and fraud flags.

• Datadog Cloud SIEM default rule: security notification rule modified or deleted

  Runtime detection that alerts when someone edits or deletes a security notification rule, which could be an attempt to silence alerts. Datadog Monitor Quality also flags monitors muted for more than 60 days. Both partly overlap the drift-report idea, but only inside Datadog.

• Kennel / Terraform Datadog provider

  Monitors-as-code tools. Running terraform plan shows drift between the live monitors and the code. They are the DIY building blocks for a monitoring drift report, but they produce no posture-level report showing what was weakened.

Checked 2026-10-21