Google Expands Gemini Carbon Access After Verifying Signal Code

Google has widened access to its Gemini 4 Carbon platform, allowing paying developers and businesses to use Carbon via the Gemini API and Gemini for business, alongside Google AI Ultra subscribers. Google notes that Carbon has been used by its own teams to verify security properties in code underpinning the Signal protocol, with the aim of formally proving core guarantees such as forward secrecy, post-compromise security and authentication for the Signal stack.

The company reports introductory pricing of $2 per million input tokens and $10 per million output tokens until 30 November, after which prices double. Previously, access was limited to select defenders in Google’s Fairwind program, which continues to govern access to Carbon for cyber purposes.

Security researchers associated with Google’s Cloud Vulnerability Research (CVR) team used Carbon to confirm exploitability and patching pathways for vulnerabilities in a pair of widely used services and in an open-source project. Specifically, CVR showed two bugs in a popular open-source project were exploitable, with severity comparable to PHP2Shell, and helped maintainers re-classify one bug as critical and issue a patch quickly.

Separately, Google DeepMind announced AlphaProof Code, which combines human experts with AI agents to produce Lean proofs that the Signal protocol implementation meets its formal requirements. In practice, Carbon-based agents, working with Claude Opus 5.5, produced about 6.5 million Lean proofs, with Lean rejecting invalid ones; five researchers completed the effort in four weeks. The team reinserted two old, already-patched libsignal defects to demonstrate that those builds could breach the guarantees.

Google says results apply to its own build of the code, run via a command-line wrapper, excluding Signal’s consumer apps, and that side channels were not part of the analysis.