AI startups Intology and Autoscience submitted AI-generated studies at a conference without disclosure and face criticism of co-opting peer review for publicity
Kyle Wiggers / TechCrunch : X: @intologyai , @pandaashwinee , @intologyai , @tuhinchakr , @sakanaailabs , @autoscienceai , @autoscienceai , and @dorialexander X: @intologyai : Zochi's papers received unanimously positive reviews: 1️⃣ CS-ReFT received strong peer review scores (6,7,6), with reviewers commending its “clever idea” and effectiveness in addressing “a critical limitation of ReFT.” 2️⃣ Reviewers gave Siege scores of (7,7), highlighting the @pandaashwinee : i think submitting ai papers to a venue without contacting the PCs is bad. Sakana reached out asking whether we would be willing to participate in their experiment for the workshop i'm organizing at ICLR, and i (we) said no. this shows a lack of respect for human reviewers time. @intologyai : 🤖🔬Today we are debuting Zochi, the world's first Artificial Scientist with state-of-the-art contributions accepted in ICLR 2025 workshops. Unlike existing systems, Zochi autonomously tackles some of the most challenging problems in AI, producing novel contributions in [image] Tuhin Chakrabarty / @tuhinchakr : This is disgusting and shameful. Reviewers and Editors are already tired and overworked. Submitting low quality AI generated papers only makes things worse. There needs to be proper steps to ensure this never happens @sakanaailabs : The AI Scientist Generates its First Peer-Reviewed Scientific Publication We're proud to announce that a paper produced by The AI Scientist-v2 passed the peer-review process at a workshop in ICLR, a top AI conference. Read more about this experiment → https://sakana.ai/... [image] @autoscienceai : Introducing Carl, the first AI system to create a research paper that passes peer review. Carl's work was just accepted at an @ICLR_conf workshop on the Tiny Papers track. Carl forms new research hypotheses, tests them & writes up results. Learn more: https://autoscience.ai/... @autoscienceai : At Autoscience, we're solving autonomous AI research. We just released a technical report on Carl, the first AI system to create an end-to-end AI research paper accepted past peer review. 🧵 [image] Alexander Doria / @dorialexander : I was half-joking with Sakana but this needs to stop immediately. Academia is not there to outsource free LLM evals. Get the f**ck out.
Context & Ripple Effects
The dispute extends a longstanding AI-research transparency problem: earlier coverage found that reproducibility was already constrained by missing code, data and other access barriers, while journals were later forced to address undisclosed generative-AI help in manuscripts. Earlier evidence of missing research artifacts made it harder to independently assess AI claims even before automated paper production became a concern.
This case shifts the focus from authorship assistance to the use of conference review itself as a public signal for AI systems. It also lands in a field where peer reviewers were already carrying much of the ethical-oversight burden, making undisclosed machine-generated submissions especially contentious.
First-order effects
- Intology and Autoscience face credibility damage over publicizing workshop-review outcomes without clearly disclosing that Zochi and Carl generated the submitted work; the affected ICLR workshop and its reviewers must contend with whether their process was used under understood terms.
- The episode puts immediate pressure on conference organizers to clarify disclosure and consent expectations for AI-generated submissions, rather than leaving reviewers to discover the provenance after the fact.
Second-order effects
- Other AI-paper-generation startups may find it harder to treat acceptance or reviewer scores as straightforward product validation, particularly when the review pool did not agree to evaluate a system experiment.
- Program committees may add provenance questions, AI-use declarations, or separate experimental tracks, increasing submission-screening work but reducing reviewers' exposure to undeclared evaluations. This follows the same governance gap exposed when journals began confronting undisclosed generative-AI drafting.
Third-order effects
- If automated research agents become routine submitters, peer review could split more clearly between evaluating scientific contributions and benchmarking AI systems; each requires different consent, disclosure and evaluation rules.
- The structural risk is that scarce reviewer labor becomes an unpriced input to AI-product marketing. Durable norms will likely depend on venues making provenance and experimental participation explicit, not merely on post-publication criticism.
The trend: AI research institutions are moving from broad concern about generative-AI authorship toward operational rules for how autonomous systems may enter, and be evaluated by, scholarly processes.