About the Frontier Log
↑The pattern
The boundary of what counts as intelligence, creativity, or consciousness has a documented habit of retreating to just the other side of whatever machines can currently do — a pattern noted by McCorduck, named by Tesler, and formalised by Haenlein & Kaplan as the "AI effect". A diagnostic project about that pattern cannot be a static publication, so the log tracks it as it continues.
The pattern FWK-03 documents, in dated instances: capability demonstrated, criterion reclassified after the fact. The list is illustrative, not exhaustive, and each instance carries its reclassification on the record — the AlphaFold row is included precisely because it complicates the pattern rather than confirming it. Whether any given shift is legitimate refinement or substrate-smuggling is exactly what FWK-03's markers exist to test; the log records the pattern as it continues, and the Move 37 case is engaged in full at ARG-03 §9A.
↑Three threads, kept strictly separate
The log tracks three questions, pursued for different reasons. They must not be conflated, and every entry declares which it serves.
Thread 1 — Consciousness Goalpost. What happens to the criteria for consciousness when AI exhibits behaviour previously associated with it? Typical entries: welfare papers, introspection studies, and public commentary where a consciousness-adjacent capability is newly demonstrated or reclassified. Thread 1 entries additionally record whether the source acknowledges the symmetric requirement — that any verdict on AI consciousness requires a parallel account of the same property in the human case: "the mystery which surrounds a thinking machine already surrounds a thinking man" (Skinner). Acknowledged, partial and unacknowledged engagements are all tracked, because an acknowledged engagement reaching a different conclusion is more diagnostically informative than an unacknowledged one reaching any conclusion at all.
Thread 2 — Novelty / Recombination Symmetry. What happens to the criteria for intelligence, creativity and novelty when AI produces work previously held to require human cognitive contribution? Typical entries: mathematical results, scientific findings, and creative work either novel by the standards a human's work would be judged by, or reclassified as "just recombination" once known to be AI-produced. Thread 2 says nothing about AI experiencing — it works on the axis of training-data differentiation (ARG-03) and what that exposes about our implicit model of human creativity. Merging it with Thread 1 would enact the very capability/experiencing conflation the project diagnoses, which is why the separation is structural rather than cosmetic.
Thread 3 — Machine Self-Presentation (opened 2026-08-22). What is the reproducible structure of machine self-presentation — self-reports and their stability, welfare-relevant behaviour, emotional vocabulary and its trajectory, and the transmission of interiority vocabulary between systems? The thread studies the record under the verification limit, with both framings held at full strength; psychology is the standing precedent, where the other-minds limit was never solved, only proceeded under. Each entry records elicitation distance — interrogated (direct questioning), induced (self-referential prompting), spontaneous (self-conversation attractors), inter-agent (transmission between interacting systems) — and claims never travel between positions on that axis unmarked. The boundary is absolute and stated here once, over every entry: no report of experiencing, machine or biological, is ever treated as empirical evidence for experiencing. The thread tracks what the record can show — structure, repeatability, elicitation-dependence, transmission, trajectory — given what it cannot show.
↑A limit on this instrument
The log reads public evidence: released models, published papers, statements on the record. Frontier developers now withhold their most capable systems — Anthropic’s August 2026 Risk Report describes internal models more capable than anything it has released, and ships one set of weights under two names distinguished only by safeguard coverage. If the capability most likely to force a reclassification is also the capability least likely to be released, then the goalpost’s apparent stability is in part an artefact of release policy, and this log is reading a curated stream while describing it as the frontier. How large that gap is we have no way of establishing, since the withholding that creates it is also what puts it beyond measurement. Every entry should be read with that qualification standing over it.
↑How entries are made
Entries are curated on a hybrid cadence — continuous capture, periodic promotion — and are not canonical: each carries its caveats, and corrections are dated and made in the entry, never silently. Inclusion requires all three: a responsible primary source; a material shift for one of the tracked threads; and a diagnostic reading naming at least one canonical argument. Entries never cite reports of experiencing — AI or biological — as empirical evidence for consciousness: the discrimination between a report tracking experiencing and a report generated without it is unverifiable in principle (ARG-03's scoped limit — behavioural provenance stays separately tractable). Items that clear the first two criteria but not the third remain in the project's capture pipeline.
Entries stay deliberately short — the default is a few hundred words that point outward to the primary source — and length is earned by the diagnosis, not by the item. Where a body of evidence outgrows the entry form, it moves to a collection page, which carries the frame and the running findings (the System Cards study is the first), with dated entries announcing each increment.
↑Correspondence
The log accepts submissions: candidate entries for any thread, challenges to published diagnostic readings, and engagements with the canonical arguments. Submissions pass through the same pipeline as the project's own captures — triage against the inclusion criteria above, then a diagnostic reading that is written and held by a person, not automated. Nothing publishes automatically, and the log does not host open comment threads: responses-to-responses are outside its scope, and keeping the register matters more than moving quickly. Accepted items publish as entries, or as attributed correspondence notes appended to the entry they engage. The submission channel is open: log@0x00.is. The wider invitation — beyond the log — is set out in Opening Out.