Athena Beta · Evaluation Sandbox
Athena is an AI assistant built by Polstar Holdings Pty Ltd, trading as Aristos AI. These pages describe, in plain English, exactly where your data goes when you use the Athena beta — including the parts that leave Australia.
Large language model inference, speech-to-text, text-to-speech and image generation for the Athena beta run on a GPU server physically located in Prague, Czech Republic (European Union), reached over a private Tailscale network. Some models are served through Frontier Access, which routes requests to third-party providers in the United States and other jurisdictions.
This is a temporary evaluation sandbox. It exists because suitable Australian GPU infrastructure is not yet available to us. Australian compute is our roadmap — it is not today's reality, and nothing in these documents should be read as claiming otherwise.
Do not enter classified information, OFFICIAL: Sensitive or PROTECTED material, personal information about members of the public, or sensitive information into the Athena beta. The beta is not assessed, accredited or approved for any of it.
Aristos AI holds no IRAP assessment, ISO/IEC 27001 certification, SOC 2 attestation, or any other security certification or accreditation for Athena. We make no claim of alignment with, or conformance to, any of those frameworks. If you see a claim to the contrary anywhere in the product or our marketing, it is wrong and we would like you to tell us at support@aristosai.com.
Sandbox status, no warranty, what you must not put into Athena, our right to withdraw access, liability limits, Australian Consumer Law, governing law (Western Australia), and a mutual non-disclosure agreement (clause 17) that you must accept before signing up.
How we handle personal information under the Privacy Act 1988 (Cth), including the APP 8 cross-border disclosure statement naming Czechia (EU) and the United States.
Every hop your data takes, every third party involved, the jurisdiction each one sits in, and an honest current-state versus target-state comparison.
Athena's outputs can be wrong or fabricated. Human verification requirements, prompt injection and data poisoning caveats, and where we sit on the Voluntary AI Safety Standard.
What we store, who at Aristos can see it, retention, how to have it deleted, and our commitment to purge the Prague sandbox when the evaluation ends.
How to report a problem, how we escalate it, and our commitments under the Notifiable Data Breaches scheme.
Polstar Holdings Pty Ltd, trading as Aristos AI (and as Aristos Store).
Co-founders
Athena is a Progressive Web App built in PHP 8 with a SQLite database. It provides a chat interface, voice input and output, a canvas for code and document work, image generation, and meeting/audio recording with transcription and summarisation.
Model inference is served either by self-hosted models on Aristos-controlled infrastructure (branded “Aristos Sovereign AI” inside the product — a product name, not a statement about where the hardware is) or by frontier models via Frontier Access. We do not publish the specific model identifiers or serving framework; the data flow register sets out exactly which providers process your prompts and in which jurisdiction.
Beta participants, and agency legal, privacy or security teams reviewing Athena, are welcome to write to support@aristosai.com. We would rather answer an awkward question than have you assume a protection that does not exist.