
UK consultation response · November 2026
Making agentic AI decisions open to meaningful challenge
An opportunity to appeal means little if a person cannot investigate the grounds of the decision. On November 13, 2026, I submitted a response to the Information Commissioner’s Office’s call for evidence on agentic AI. It examines what makes that investigation possible, and how its findings can reach someone with authority to act.
The response develops a hypothetical credit example. One agent retrieves payroll data, another converts it into monthly income, and a third recommends refusal. A loan officer approves the recommendation. If a four-week earnings figure has been treated as a calendar month, what would let the borrower recognize the mismatch, establish what happened, and obtain a reasoned reconsideration?
The Necessary Informational Conditions framework organizes that inquiry around the work that needs to be done. Records must let the person or an authorized representative assess a material objection. Review must be able to act on the findings. A corrected calculation may change the credit assessment, while the final decision may still be justified on other grounds. The person should learn why.
The response addresses meaningful human involvement, safeguards, logging, permissions, and interface design. It also proposes omission and substitution tests: remove a candidate informational operation and examine whether the inquiry can still be completed adequately; then test whether another arrangement can perform the same work. These are proposed tests of the framework, not reported results from a completed study.
Recent research on published contestability guarantees and regulatory developments in employment and credit provide context. The response distinguishes those sources from its own design recommendations and keeps the UK and EU legal frameworks separate.
Read the response: Download the consultation response (PDF, 6 pages)
Related work: NIC research program · ICO call for evidence