Orbital Matter tests in-orbit 3D printing with half-meter beam
The German startup also deployed a flexible solar array during the demonstration, although one of its printers failed.
Anthropic works on cryptographic checks for AI model identity
The FlashProve research project would allow users to check a model’s identity without exposing its private internal parameters.
Anthropic is researching an open-source tool called FlashProve that would use cryptography to authenticate the AI model behind a response, the company said in a November 3, 2026, blog post. Researchers have published technical documentation and code for several components, but deployment is not planned.
The researchers said the proposed checks could expose undisclosed model substitutions, including those made by the AI provider itself. Such substitutions might involve using a less expensive model to lower costs or a pricier alternative to make performance results look better.
The approach requires laboratories to publish a digital fingerprint for each released model. Users could then cryptographically check outputs against that identifier. According to the researchers, the fingerprints would leave model weights—the internal numerical parameters determining behavior—private, making the approach compatible with commercial models. Checks are intended to run locally on a user’s device.
Another potential application is testing outside parties’ assertions about their use of AI. Examples cited by the researchers include claims that a particular model checked software for vulnerabilities or approved a business plan.
The documentation and component code are available through GitHub with open-source licensing. Researchers characterized the work as preliminary and invited others to develop it further. Neither Anthropic nor another laboratory has plans to deploy FlashProve.