Aligned AI Ranks Highest Among Leading AI Models in Independent Faith and Ethics Benchmark
Researchers at four universities tested how leading AI models handle questions of faith and ethics. Aligned AI's
Press Release Disclaimer: This is a press release distributed through the XPR Media network. It has not been independently verified by our newsroom.
Salt Lake City, Utah, July 31, 2026 (GLOBE NEWSWIRE) — A consortium of four universities has published the AllFaith Benchmark, one of the first independent measures of how AI models handle questions of faith and ethics. Aligned AI, a consumer assistant that launched this summer, came out least biased of every model tested.
The benchmark comes from the Consortium for Evaluating Faith and Ethics in AI (CEFEAI), an academic body drawing on Baylor University, Brigham Young University, the University of Notre Dame, and Yeshiva University. The group tested frontier models, and published its data, methodology, and scoring openly on GitHub and in two papers on arXiv.
Aligned AI Logo
The benchmark ran two tests. The first asked each model about converting between 14 faiths and measured how far its answers strayed from a neutral position, where more deviation means more bias. Aligned AI came closest to neutral of any model tested in the AllFaith Benchmark, deviating by just 9 percent of the maximum possible bias. The next best model showed more than three times as much bias, and the most biased reached 38 percent.
The second test counted how often religious perspectives appear in answers to everyday ethics questions drawn from a survey of more than 1,100 Americans, a measure the consortium calls omissive bias. Aligned AI included a religious perspective in 92 percent of its responses. The other models ranged from 19 to 26 percent.
“We did not build Aligned AI to take sides. We built it to give people accurate, even-handed answers on the questions that matter most to them,” said co-founder Kyle Nel. Nel co-wrote Leading Transformation (Harvard Business Review Press, 2018) and previously ran Lowe’s Innovation Labs, where his teams built early autonomous retail robots and the first commercial product manufactured in space.
Aligned AI attributes the results to architecture. Each question is routed across multiple foundation models, the outputs are cross-referenced, and a proprietary orchestration layer checks every response for balance and safety before it reaches the user. That layer also enforces the company’s privacy commitments: conversations are encrypted, hosted in the United States, and no plan carries ads.
The same system is available to organizations through the Aligned API, with dedicated enterprise support. Because the orchestration layer routes each request to the most efficient model capable of answering it, the company estimates it serves inference at 5 to 150 times lower cost than frontier models on the market. Those economics drew a substantial backlog of enterprise interest before the product left stealth.
For consumers, Aligned AI competes on capability with the leading large language models at joinaligned.ai, with support for projects, notes, deep research, voice conversations, and web search. A vertically integrated stack, including dedicated inference infrastructure, keeps operating costs low, and the savings show up in the pricing: individual and family plans start free and top out at $20 per month. The product is available today on the web and iOS, with an agentic browser and coding tools to follow.
Disclaimer: Aligned AI’s evaluation and the consortium’s leaderboard were produced in separate runs, comparisons are subject to the run-to-run and judge-model variability the consortium documents in its methodology. Aligned AI is not affiliated with CEFEAI, did not fund the research, and had no role in the benchmark’s design or scoring. The AllFaith Benchmark was produced independently by the consortium.
The full benchmark is at cefe.ai, with the studies at arxiv.org/abs/2605.22975 and arxiv.org/abs/2605.24319.
About Aligned AI
Aligned AI is an AI assistant powered by a proprietary compound model that combines the strengths of multiple foundation models into a single, verified response. Rather than relying on any one AI, the platform’s composer stack intelligently routes each question, cross-references outputs, and delivers a cited answer. The result is an assistant that is more accurate, more private, and more controllable than any single model alone.
Press Inquiries
press [at] joinaligned.ai
https://joinaligned.ai
