The engineer three desks over has cloud admin tokens, registry publish rights, a production kubeconfig, and API keys for four AI providers cached on a laptop where postinstall scripts run automatically. Whether that engineer works at a large firm or a five-person startup, that machine is the most privileged and least governed node in the software supply chain. The 2026 campaign wave proved attackers know it.

This year alone: a state actor took over an entire npm scope and published 140+ packages that executed at install time. A worm’s source code went public and spawned copycat campaigns within weeks. Malicious packages shipped with valid SLSA provenance and sailed straight past the attestation checks teams were told to trust. A security vendor’s own CLI package was hijacked to harvest cloud, wallet, and AI-assistant credentials from developer machines. Roughly nine in ten supply chain campaigns now touch the developer endpoint. Even npm conceded the fight. Version 12 blocks install scripts, Git dependencies, and remote sources by default; package installation is moving from implicit trust to explicit allowlisting, and the endpoint inherits whatever the registry can no longer catch.

This talk applies Zero Trust tactics to the engineering workstation without wrecking developer velocity. We map the local toolchain as a protect surface (data, applications, assets, services), then walk eight defensive techniques: credential scoping and short-lived tokens, egress control, execution allowlisting, dependency cooldowns and release-age gating, lockfile and provenance verification, IDE extension governance, AI agent skill vetting, and contextual data loss prevention. Every technique maps to OWASP SPVS and NIST SSDF controls and sits on a three-level maturity spectrum, from the solo developer with no IT department to the fleet-managed enterprise.

Come for the breach stories. Leave with the Monday checklist your platform team will actually adopt.