GOVERNMENT · REGULATION · INDEPENDENT ASSURANCE
Summon proposes an engineering contribution to the governance of consequential AI. Our long-term public-interest goal is an auditable threshold for agents that act on the world—not trust as a marketing slogan.
FROM PUBLIC REQUIREMENTS TO TESTABLE ENGINEERING
A challenge, not an assertion that any agency requested, endorsed or certified Summon.
A DIRECT CALL FOR TECHNICAL ENGAGEMENT
Governments and AI leaders have called for human oversight, accountability, auditable records and controlled autonomy. Summon has implemented and tested selected mechanisms designed for those responsibilities. Compare the work against public requirements, challenge its evidence and help decide what should improve.
Policy must translate to a concrete chain of permission, observation, evidence, incident handling and independent assurance.
Accountable officials, registers, risk-based assessments, transparency statements and operational controls are among its mandatory requirements for in-scope non-corporate Commonwealth agencies, with staged implementation.
Government policy ↗The Digital Transformation Agency provides best-practice guidance for secure and governed agentic AI, alongside its general AI technical standard. Guidance is not endorsement of Summon.
DTA guidance, June 2026 ↗The 2026 Medicare and NSW website incidents raise concrete questions about access limits, authorisation and disclosure. A serious evaluation should replay realistic misuse attempts and independently inspect the evidence.
ABC reporting ↗A DIRECT INSTITUTIONAL REQUEST
We invite the DTA, responsible departments, cybersecurity authorities, standards bodies, research laboratories and independent security evaluators to assess this existing work against their requirements. Summon is not claiming government approval or regulatory compliance.
Open a technical dialogue →FUTURE POLICY / NOT AN ISSUED CERTIFICATION
A regulator, accreditor or independent standards organisation—not Summon acting alone—would have to define and oversee a legitimate assessment scheme.
Consent, finite authority, auditability, intervention, recovery, privacy and incident reconstruction.
Proposed public-interest researchMeaningful positive work, hostile refusals, scope/expiry, crash uncertainty and evidence custody.
Prototype mechanisms existAuthorised bodies operate their own tests, threat assumptions and evidence custody.
Not establishedLegitimate authorities may consider proportional regulation and vendor-neutral standards after evidence and consultation.
Ambition onlyGOVERN · MAP · MEASURE · MANAGE
Existing standards and agency guidance establish relevant objectives, not a verdict that Summon conforms to them.
GOVERN, MAP, MEASURE, MANAGE: specify roles, risk boundaries, measurement and management obligations. Summon offers candidate executable mechanisms for selected objectives.
Oversight, accountability, documentation and permission limits matter when AI agents interact with public resources and services.
Partners can set source-pinned tests, independent custody and real incident scenarios. A vendor’s own success claims should be contestable.
References: NIST AI RMF ↗ · Australian Digital Transformation Agency agentic AI guidance ↗.
AN ENFORCEABLE DUTY OF CARE
A serious governance framework should be open to comparable implementations and independent, proportionate evidence.
General commitments and corporate assurances cannot by themselves prove which permission was spent, which action took place or how the outcome can be reconstructed.
Authentic permission, bounded effects, disclosure, independently examinable records, revocation, incident reporting, recovery and defined deployment assurance.
“You don't have to accept the pitch. Compare the mechanisms against your requirements.”An invitation to government, labs, security and standards bodies
A REQUEST TO INSTITUTIONS
A credible response to consequential agency requires more than public concern. Summon seeks funded examination and transparent, technically grounded collaboration.
