Specification Induced Speculation Whitepaper
Year: 2026
Type: article
Abstract: Specification Induced Speculation is the failure mode in which an AI model, given an instruction that does not fully define the intended behaviour of the system being built, fills the undefined regions with outputs that are plausible, coherent, internally consistent, and wrong — not because the model malfunctioned, but because the instruction was incomplete. The model did exactly what it was designed to do: it completed what it was given. The decisions it made in the unspecified regions were not authorised by any human, were not verified against any requirement, and are invisible in the output because they look indistinguishable from the decisions that were authorised. The failure is not hallucination — the model invented nothing. It speculated, faithfully and silently, about what was intended. The speculation was induced by the specification — or the absence of one.
Author: Abdullah, Hamza
Language: English
Subfield: Safety Research
Field: Social Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7160776593
FWCI:
Cites:
Cited by:
Related to: