Concordia AI

Theory of Change

Concordia AI's theory of change rests on a single core claim: the false narrative that "China doesn't care about AI safety" is being used to justify reckless AI development in the West, and correcting this narrative can reduce global AI risk.

Brian Tse, TIME op-ed (August 2025): "'China doesn't care about AI safety -- so why should we?' This flawed logic pervades U.S. policy and tech circles, offering cover for a reckless race to the bottom."

The mechanism: Concordia (1) produces English-language intelligence on China's AI safety landscape to reduce information asymmetry, (2) consults Chinese AI labs on safety best practices to improve actual safety, (3) convenes international forums to build dialogue and consensus, and (4) participates in Chinese national standards development to embed safety into regulation.

Brian Tse, Carnegie Council podcast: "In China AI safety and governance experts follow what is happening with the EU AI Act or superalignment work at OpenAI on a daily basis. However, their international counterparts often know very little about relevant work in China."

What They Do

Reports: Published the "State of AI Safety in China" series (2023, 2024, 2025) -- 100-150 page comprehensive analyses of Chinese domestic governance, technical safety research, industry practices, expert views, and public opinion. Cited by Bloomberg, Wired, Nature News, People's Daily. Andrew Yao (Turing Award winner) cited the series publicly. Also published the first "State of AI Safety in Singapore" (2025).

Frameworks: Co-published the Frontier AI Risk Management Framework v1.0 with Shanghai AI Lab at WAIC 2025 -- China's first comprehensive framework for managing severe risks from frontier AI. Defines "red lines" and "yellow lines" across cyber offense, biological threats, persuasion, and loss of control. Jack Clark (Anthropic co-founder) noted convergence with Western risk evaluations. Accompanying technical report evaluated 20+ models across 7 risk domains.

Monitoring Platform: Launched airiskmonitor.net tracking 34+ frontier LLMs from 15 developers on cyber, biological, chemical, and loss-of-control benchmarks. Q3 2025 findings: risk indices at record highs across all domains, GPT and Claude families maintain "stable low risk."

Standards Participation: Member of SAC/TC260, SAC/TC28/SC42, MIIT/TC1, and AIIA Safety Governance Committee. Deputy chief expert at AIIA. Contributed to multiple national and local AI safety standards including Shenzhen's "Technical Framework for Value Alignment of Pre-trained AI Models."

Convenings: Hosts AI Safety and Governance Forum at WAIC (China's largest AI conference). 2023 speakers included Hinton, Altman, Russell, Yao. 2025 speakers included Bengio, Russell, UN Under-Secretary-General, PAI CEO, 30+ others. Co-hosted events at Paris AI Action Summit, BAAI Conference, ICLR Singapore.

Summits: Bletchley Park (2023, closing plenary speaker -- one of four non-governmental Chinese attendees), Seoul (2024, one of 10 civil society orgs), Paris (2025, Chinese civil society representative). Also IDAIS Shanghai Consensus signatory, NTI AIxBio Statement signatory.

Consulting: "Signed strategic partnership agreements with several leading Chinese general-purpose AI developers." Services: regulatory compliance consulting, automated red-teaming, advisory briefings. Helps Chinese labs navigate international regulations for market entry. Clients are unnamed.

Newsletter: "AI Safety in China" Substack (1,400+ subscribers, 24+ issues), WeChat (4,600+ subscribers). Chinese Perspectives on AI Safety website (chineseperspectives.ai).

Key People

Brian Tse (Founder & CEO): HKU grad, started in deep learning hardware at Tsinghua ~10 years ago. GovAI Policy Affiliate (Oxford). Former PAI Senior Advisor. Consulted OpenAI on GPT-2 release (2019). Co-founded EA Philippines. Translated 5 major AI safety books into Chinese (Ord, Russell, Tegmark, Christian, Fu Ying). His personal network -- spanning GovAI, PAI, OpenAI, BAAI, Tsinghua, Carnegie, UN -- is the organization's primary asset.

Kwan Yee Ng (Head of International AI Governance): Writer for the International AI Safety Report (Bengio-led). Research affiliate at Oxford Martin AI Governance Initiative. Yenching Scholar at Peking University. The most prominent Concordia voice after Tse.

Team structure (14-15 staff, mid-2025): Deliberate mix of Chinese policy insiders (ex-Baidu technical consultant Liang Fang, ex-Baidu R&D engineer Weibing Wang) and Western-educated researchers (Cambridge, Oxford, Harvard, LSE). Ex-ByteDance TikTok policy (Yuan Cheng). FLI Fellow with CVPR/ICML/NeurIPS publications (Yawen Duan). Ex-Singapore government (Jonathan Lee). Growth from 5 staff in 2023. 100+ affiliates across 7 cohorts (names and institutions not public).

Money and Incentives

Revenue: EUR 6.12M (2024, per Tracxn). Zero external funding raised. No grants from Open Philanthropy, SFF, EA Funds, or any other identified philanthropic funder. Revenue appears entirely from consulting to Chinese AI labs.

Business model: Consulting to Chinese AI developers on safety compliance, red-teaming, and international market entry. This is the primary revenue source. The social enterprise certification (Beijing Civil Affairs Bureau, 2023) commits at least 35% of after-tax profits to public purpose projects.

Clients: Unnamed. Concordia describes "strategic partnership agreements with several leading Chinese general-purpose AI developers" but has never identified them. Given staff backgrounds (ex-Baidu, ex-ByteDance) and WAIC forum speakers, likely clients include major Chinese labs.

Structural conflict of interest: Concordia's revenue comes from consulting to Chinese AI labs on safety. Concordia also produces the definitive English-language reports assessing how seriously China takes AI safety. These activities create an incentive for Concordia to present Chinese AI safety efforts favorably. No external board, no published conflict-of-interest policy, no editorial independence safeguards are disclosed.

Consulting-to-standards pipeline: Concordia sits on national standards committees (TC260, AIIA) that define what AI safety compliance means in China, while also consulting for the labs that must comply with those standards.

Salary range: US$68,000-108,000 (Fall 2025 job postings).

Comparison to peers: Most AI safety organizations are heavily grant-dependent. Concordia is entirely self-funded through consulting revenue. This grants independence from Western philanthropic priorities but creates financial dependence on Chinese AI industry goodwill.

What Others Say

"The False Case for Cooperation With China" (Robert Atkinson, ITIF/PolicyArena, July 2025): "The CCP is realist to the core, caring only about China. When it sees the U.S. government requesting cooperation, it sees leverage." Argues AI cooperation with China is unnecessary because "The United States doesn't need Chinese input to regulate AI domestically." Not Concordia-specific but directly challenges the cooperation paradigm Concordia embodies.

China Media Project (July 2025): Chinese AI safety standards encode CCP information control. The 31 "Main Safety and Security Risks" put "core socialist values" violations at the top. Tests showed Chinese chatbots (including Zhipu, whose representatives attend international safety gatherings) stating "there is no such thing as a 'democratization process' in Taiwan." Two scientists who co-authored an international AI safety paper alongside Carnegie researchers were also involved in drafting these politically-oriented standards.

Carnegie Endowment (Sheehan, August 2024): "Despite the growing salience of safety concerns, China's leaders remain just as, if not more, worried about falling further behind the United States in advanced AI." Chinese policy advisers invoke: "Failing to develop is the greatest threat to security."

AI Frontiers (October 2025): "The AI safety ecosystem in China remains far less mature than its US counterpart in both technical and policy domains." CnAISDA "has so far taken little substantive action to address potentially global-scale risks." DeepSeek was "conspicuously absent" from the second round of voluntary safety commitments.

Brian Tse himself (Cognitive Revolution podcast): Chinese companies' safety practices "appear to be primarily geared towards the 31 categories of risk there are in the existing regulation and the vast majority of Chinese companies have not publicly shared evaluation results for CBRN or loss of control risk."

Zero direct criticism of Concordia AI exists in publicly searchable sources. No published critique specifically targets Concordia's independence, accuracy, or value. The AI safety community (LessWrong, Alignment Forum, EA Forum) has never discussed Concordia.

What's Absent

No external board or governance structure disclosed for an org with EUR 6M+ revenue
No named consulting clients despite advocating for international transparency
No acknowledgment of how Chinese "AI safety" standards encode political censorship alongside genuine safety
No editorial independence policy explaining how Beijing-based reporting navigates CCP constraints
No financial audit or detailed revenue breakdown beyond Tracxn estimate
No methodology for handling politically sensitive findings in State of AI Safety reports
No discussion of Chinese companies' failure to fulfill Seoul AI Safety Commitments in Concordia's public communications
Affiliates program (100+ fellows) has no public names or institutional affiliations