01
The record
Arkna records the AI activity it receives and prepares it for later review.
About Arkna
Arkna keeps a readable record of the AI activity it receives. It is prepared for someone who has to examine an action later, and it states what the record covered and where evidence may be missing.
ARKNA PTY LTD is an Australian company based in Sydney. We build for institutions that need to account for machine action after the moment has passed.
Our mission
Machine action open to human examination.
Oversight of these systems has to stay with people. Machines are taking on decisions about credit, claims, entitlements and care, and the room to ask what happened has to remain. What a machine did inside a black box should be knowable from outside it. The system cannot be the sole author of its own account.
Read the letter →Why Arkna exists
Financial institutions are putting agents into production faster than they can account for them.
The executive responsible for an agent process may still be unable to answer four questions about it: what it did, on what basis, who approved it, and whether the account is complete.
The reason is structural. The systems that take the action also produce the account of it, and visibility degrades on the way up. The deployment team sees the most, the CIO sees less, the CRO sees a summary, and the board sees a slide. By the time a question arrives from outside the firm, the account is several removes from the event.
Arkna closes that distance with a record made as the activity happens rather than assembled afterwards. The institution owns the record and decides who can examine it.
What we do
02
Research
We publish on whether machine action can be examined after the fact, and by whom. The work uses primary sources and states the limits of what they establish.
What the record makes possible
01
Review grounded in the run
A reviewer can examine the action itself rather than rely only on a policy, dashboard or account produced later. The institution decides what the record establishes and what should follow.
02
Reliance across suppliers
AI systems rely on outside services, and those providers may rely on others. Arkna identifies the model and version observed on the paths it covers, and states where that coverage ends.
Capability and limits
Arkna gives people evidence they can examine after an AI system acts. It does not run the system, approve its actions or decide whether the outcome was fair, safe or compliant. Those responsibilities remain with the institution.
Every record says which activity Arkna received. It can show whether that material was changed later, but it cannot account for an action it never saw. The record supports monitoring and audit; it does not replace them.
Keith Cheng
Founder
In our lifetime, machines may act with more agency than we do. Whatever they become, I want them to carry a record, so that what happened stays something a person can look at.