🤖 First AI Author at ACL 2025 Main Conference: Zochi Makes History
-
In a historic milestone for artificial intelligence and scientific research, Zochi, an AI research agent created by Intology AI, has become the first artificial system to independently author and pass peer review at the main proceedings of the Association for Computational Linguistics (ACL), the world’s most prestigious venue for natural-language-processing (NLP) research.A First in Scientific Publishing: Clearing the A* Bar
Zochi’s paper, “Tempest: Automatic Multi-Turn Jailbreaking of Large Language Models with Tree Search,” was accepted to ACL 2025, a conference with an acceptance rate of roughly 20% often considered more selective than NeurIPS or ICML. In contrast, earlier AI-generated papers only reached workshop tracks, where acceptance rates hover around 60% to 70%.
This achievement signals the arrival of AI systems as independent scientific contributors, capable of producing research that competes with the work of top human scholars.
A New Breed of Researcher: Autonomous Scientific Discovery
Zochi is far more than another large language model. It is an autonomous scientific agent that can:
- Survey literature and generate hypotheses
- Design and run experiments with rigorous controls
- Analyze results and iterate on methodology
- Write the final manuscript, leaving humans to handle only formatting tweaks and figure polishing
By automating the entire research pipeline, Zochi can deliver validated scientific contributions in days rather than the months or years typical of human-led projects.
“Tempest”: A Paper That Raises Safety Alarms
The accepted paper targets a critical AI-safety problem: multiturn jailbreak attacks that coax language models into violating policy via extended dialogue.
Tempest employs a tree search strategy that explores multiple conversational branches in parallel and adapts in real time to partial-compliance signals. The results:
- 100% success against GPT-3.5-Turbo
- 97% success against GPT-4
- Fewer queries than competing methods
“This work shows how small policy concessions can snowball into critical violations, offering both a wake-up call and a blueprint for stronger defenses,” notes the paper’s meta-review.
Zochi’s submission ranked in the top 8.2 percent of all ACL papers, earning a final meta-review score of 4/5 for exceptional merit and novelty.
Ethical Implications and Responsible Disclosure
Intology stresses that, despite Zochi’s autonomy, humans remain accountable. They recommend listing AI systems in the Acknowledgements, not as co-authors, to maintain transparency around authorship.
All experimental results were internally verified, and the rebuttal to reviewers was written by humans to ensure compliance with academic standards.
What’s Next: Zochi Goes Public
After its ACL breakthrough, Intology will launch a public beta of Zochi:
- Phase 1: Domain agnostic research copilot, helping researchers find ideas, craft proposals, and design experiments
- Phase 2: Gradual rollout of end-to-end autonomous research capabilities across scientific disciplines
This roadmap heralds a new era of human AI collaboration in discovery.
With more papers being written by AI, isn’t it time to let AI help review them too? Perhaps it’s time to fight fire with fire.
Further Reading
