Lambda Web Functions launched on 13 October. The endpoint auth type defaults to none, one execution role serves every request, existing lambda:* grants already cover it, and posture scanners can't see it yet.

One deploy puts your Lambda Web Function in three regions behind one domain. The access model is replicated with it: slow revokes, secrets copied to every region, and audit gaps that hide who called.

While nearly every cloud team agrees on the principle of least privilege, very few actually reach it. In our latest piece, Act Security CEO Jonathan Langer breaks down why over-permissioning happens - not because engineers are careless, but because complex cloud logic makes predicting impact nearly impossible. We explore why measuring "effective access" is the missing key to shifting policy management from guesswork to testable engineering, especially as autonomous AI agents enter the environment.

The recognition spotlights Act’s access model that reasons across identity, network, and AI as a single layer and scopes every human, workload, and AI agent down to what it genuinely needs, eliminating potential access paths available for agents or attackers.

A deep dive into how Act runs every Terraform change through a self-hosted Kubernetes platform engineered for zero standing access and seamless developer workflows.

Guided, interactive tour

Jonathan Langer, CEO of Act Security, explains Act's action-centric approach to cloud access in seven minutes, then takes questions from a panel of security leaders.

One extra field on an Athena API call returned the full text of SQL queries that other AWS accounts had just run, along with the account IDs that ran them. No query results came back, but statement text carries filter values, inserted rows and table structure. We reported it to AWS on 6 August 2026 and they closed it globally in four days. No customer action is required.

Access is the foundation of cloud security, and it is also the thing engineers dread touching. The documentation manages to be both overwhelming and insufficient, the AI models are confidently wrong, and questions that sound trivial turn into afternoons of work.