Google Deepmind's AlphaProof Code verified Signal's cryptographic core in four weeks instead of years
Source: The Decoder (opens in a new tab) · Manuel Uth
Intel Summary
Google DeepMind used its AlphaProof Code system to mathematically verify Signal's cryptographic core code in four weeks, compressing a process that typically requires years. AI agents powered by Gemini Carbon generated over six million lines of formal proof code, requiring human review for approximately one percent of the output. DeepMind stated plans to apply this automated verification methodology to critical infrastructure software, including power grids and medical devices.
Why It Matters
Formal verification of complex cryptographic software has historically been constrained by extreme manual effort and specialized labor. Automating mathematical proofs at this scale significantly lowers the barrier to mathematically guaranteeing software correctness in high-assurance environments and critical infrastructure.
Organizations & Entities
Topics
Related Intelligence
- ReportAlso involving Google DeepMind
Google DeepMind just mathematically verified key parts of Signal
Ars Technica reports that Google DeepMind completed a four-week project using AI to mathematically verify key components of the Signal messaging application, demonstrating that AI tools can assist in formal verification of critical software.
Ars Technica - ReportRelated through Google
Google retools Kubernetes for AI agents instead of building a new stack
SiliconANGLE reports that Google LLC is adapting Kubernetes to support AI agent workloads rather than developing a new infrastructure stack. The shift responds to agentic operational demands, including real-time untrusted code generation, prolonged idle states, and rapid pod turnover that challenge standard Kubernetes control planes.
SiliconANGLE - ReportRelated through Google
Google's Gemini 4 Hydrogen outscores the bigger Argon just four weeks after its launch
Independent benchmarks from the Artificial Analysis Intelligence Index indicate that Google's Gemini 4 Hydrogen, which replaces the Flash series, outscores Google's larger and more expensive Argon model. The testing shows Hydrogen approaches Claude Sonnet 5.5 performance at a lower price point, though it costs roughly 2.5 times more per task than GPT-6.1 Sol and lags on agentic coding tasks.
The Decoder - ReportRelated through Google
Agents may soon be the main workload, and most of the time they’re doing nothing
Google has introduced Agent Substrate, an open-source sandboxing project designed to isolate AI agents that execute code, use tools, or control web browsers while limiting compute overhead in enterprise environments.
SiliconANGLE