DeepMind’s AlphaProof Code: agentic verification proves Signal’s core implementation

What changed: Google DeepMind published a research blog introducing AlphaProof Code, a multi-LLM, agentic framework that generated over six million lines of machine-checked Lean proofs and formally verified key security properties of Signal’s libsignal implementation in about four weeks.

Why it matters: Founders and engineering leads building security- or safety-critical systems can now consider agentic formal verification as a practical option for deep assurance — not just a research demo — because DeepMind open-sourced artifacts and describes a workflow where models do the heavy lifting and humans review one percent of the results.

Try/watch: If you operate infrastructure or security software, read DeepMind’s artifacts and the technical paper, then run a small pilot: pick a compact critical module, attempt a machine-checked proof pipeline, and measure how much human review is still required and how long integration takes.

Linux Foundation launches Agent Name Service (ANS) to give agents verifiable identities

What changed: The Linux Foundation announced that Agent Name Service (ANS) has launched as a foundation project; ANS uses domain-anchored DNS records and certificate-based signals to register, publish, and verify AI agent identities, with 11 organizations listed as early supporters.

Why it matters: For operators, integrators, and platform buyers, ANS creates a standard way to bind an agent to an owning organization and to check provenance before granting access or automations — a practical control you can require in procurement and runtime policies to reduce impersonation and supply-chain risk.

Try/watch: Start mapping which agent endpoints and third-party assistants in your stack can (or should) publish ANS identities and require them in new contracts; monitor adoption by identity and DNS providers and by agent directories that integrate ANS.