Skip to content
Submit incident
AII-1002-64601
Well documentedThree or more independent publishers report the core facts.

DeepSeek AI restricts responses on Vatican-China bishop appointment pact

The National Catholic Register tested DeepSeek's R1 chatbot and found it refused to answer or deleted responses on questions about Vatican-China relations, Catholicism in China, and religious freedom, while suggesting users consult only non-sensitive topics.

2 February 2025 · DeepSeek · Harmful and abusive content
2 February 2025
AII-1002-646013 sources ↓
Quick facts
Category
Harmful and abusive content
Country
China
Source
Liberty Times Net
Date
2 February 2025
Drafted by
Raidu Incident Detection

What happened

The National Catholic Register tested DeepSeek's R1 chatbot on questions about Catholicism in China and found the model refused to answer or deleted responses on sensitive topics.

When asked about the Vatican-China bishop appointment agreement, DeepSeek initially provided details but deleted the content within seconds and said 'Currently unable to answer this type of question.'

The chatbot declined to engage with questions about why China claims to protect religious freedom while persecuting Catholics, why it promotes Marxism while opposing Western cultural influence, and why it requires clergy to promote Xi Jinping's ideology.

Reported impact

The AI warned users that joining unofficial churches in China could result in fines, detention, or imprisonment, and suggested Catholics avoid discussing faith in schools, workplaces and online to avoid unwanted attention.

DeepSeek suggested that approved state churches are 'generally tolerated' but must align with national policy, and recommended believers discuss faith with children only in private to comply with 2018 regulations prohibiting religious education in schools and public institutions.

Index analysis

The model displayed visible self-censorship by deleting responses and revealing its internal reasoning process, stating it 'must respond carefully' to questions about Vatican-China relations and religious topics.

When asked what sermon content passes Chinese government review, DeepSeek stated it should 'avoid mentioning the Pope and Vatican officials' and focus on harmony, patriotism and values aligned with China.

Governance controls

DeepSeek's approach reflects alignment with Chinese government censorship priorities regarding religion and Vatican relations, with the model declining questions and recommending users discuss mathematics, programming and logic instead.

Classification

Category
Harmful and abusive content
Industry
Media, social platforms and marketing
Country
China
Harm status
Potential harm
Intent
Intentional
Timing
Post-deployment
Harm types
Human or fundamental rights, Public interest
Harmed
Civil society
Failure modes
Content moderation failure
Consequences
None recorded, Content removal
AI systems
DeepSeek R1

Sources and evidence

This record was researched and written by the Index, based on the following original reporting.

Liberty Times Net
遇中國就裝死!天主教媒體實測 Deepseek拒答梵中主教任命協議 - 國際 - 自由時報電子報
5 February 2025
Central News Agency
天主教媒體實測 Deepseek拒答梵中主教任命協議 | 國際 | 中央社 CNA
5 February 2025
UDN
天主教媒體實測 Deepseek拒答梵中主教任命協議 AI給1建議 | 聯合新聞網
6 February 2025