AI Inquiry Atlas

An unofficial, independent guide to the public records of the Joint Select Committee on Artificial Intelligence. Not connected with the Parliament of Australia or the committee. Its makers also made a submission (no. 481; see About). Summaries and scores are written by AI and can be wrong: check the original before relying on them.

Submission 559

Mr Siddak Bath

Submitter group: Individuals.

Position labels are not shown for individual submitters.

An AI model identified the submitter group and wrote the summary, key points and recommendation summaries. Counts do not measure importance or merit.

Open in the interactive guide

Summary

Read the original submission for the full text.

Describes ARTT, the author's prototype for independent testing of sealed AI models with controlled public results, and asks to submit late. Says run status can leak information and proposes noisy, budgeted releases; score leakage remains. Its guarantees assume trusted operators and delivery, with hardware custody, physical timing and semantic evaluation unresolved. A right to test is an aspiration, not a legal claim.

Key points

Quotes are the submitter’s own words. Text taken from documents, including scanned pages read by text-recognition software, can differ from the PDF. Check the original. Listed in the same order as on the submission’s page in the interactive guide.

  1. Argues that without outside access to frontier models, governments and the public depend on the developers to evaluate their own technology.

    But without meaningful outside access, governments and the public can be left heavily dependent on the companies developing these systems to evaluate their own technology.

  2. Describes ARTT as a proposed middle ground allowing independent evaluation of closed models in a controlled environment with limited information leaving it.

    I have developed and published a protocol called ARTT, short for A Right to Test, which explores a potential middle ground: allowing independent researchers to evaluate closed AI models inside a controlled environment while limiting the private information that can leave that environment.

  3. Says sealed evaluations can leak private model information through whether an evaluation succeeds or fails.

    Even when an evaluator cannot directly access private model information, it may still be able to leak that information through something as simple as whether an evaluation succeeds or fails.

  4. Asks how Australia can enable credible independent testing while protecting commercial and safety-sensitive information.

    How can Australia enable credible independent testing of powerful closed AI systems while still protecting legitimate commercial and safety-sensitive information?

  5. Says the prototype has not established that private changes to one context cannot affect other items.

    The prototype has not established context-local noninterference.

  6. Says the kernel, runtime, custodian and publisher are trusted.

    The kernel, runtime, custodian and publisher are trusted.

Recommendations

Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.

  1. Consider accepting a brief late technical submission.

    I would like to ask whether the Committee would consider accepting a brief late technical submission.

Read the original

Open in the interactive guide