U.S., China gear up for mid-September AI safety talks
The United States and China are preparing for bilateral talks on artificial intelligence (AI) safety, expected around mid-September, though the final agenda and list of participants remain unsettled.
The proposed dialogue is expected to focus on shared risks from frontier AI systems, including AI-enabled cyberattacks, with both sides expected to discuss frameworks for information-sharing on emerging AI risks and possible voluntary self-monitoring commitments by AI labs.
This would be among the first dedicated bilateral engagements between the two countries focused exclusively on AI safety in the current term, distinct from broader trade or technology-transfer talks.
The talks come against a backdrop of intensifying US-China technology competition, even as both governments have separately participated in multilateral AI governance efforts.
Global AI Safety Governance Architecture
Since 2023, several multilateral and plurilateral efforts have emerged to govern frontier AI risks, given the absence of a single binding international treaty on AI. The UK-hosted AI Safety Summit at Bletchley Park was the first global summit dedicated to AI safety, producing the non-binding "Bletchley Declaration."
Key Details
- The Bletchley Declaration was adopted on 1-2 November 2023, at Bletchley Park, UK, and was signed by 28 countries and the European Union, including the US, China, and India — marking rare simultaneous US-China buy-in on an AI-risk framework.
- Follow-on summits were held in Seoul (2024) and Paris/India hosted the subsequent AI summit cycle, reflecting a rotating-host model similar to climate COPs.
- India, as chair of the Global Partnership on Artificial Intelligence (GPAI) around this period, took a leadership position on "safe and trusted" AI at the Bletchley summit.
The proposed US-China bilateral talks sit within this wider architecture — even as global summits produce broad, non-binding declarations, major AI powers like the US and China are now also pursuing narrower bilateral safety dialogues to manage risks specific to their own AI competition (e.g., autonomous cyberattack capability, misuse of frontier models).
India's AI Safety Institute and the IndiaAI Mission
India has built its own domestic AI-safety governance capacity as part of a broader national AI strategy, positioning itself as a rule-shaper rather than only a rule-taker in global AI governance debates.
Key Details
- The IndiaAI Mission was approved by the Union Cabinet on 7 March 2024 with an outlay of about ₹10,372 crore over five years, structured around several pillars including "Safe & Trusted AI."
- The IndiaAI Safety Institute (AISI) was announced on 30 January 2025 under the Safe & Trusted AI pillar, following a "hub and spoke" model engaging academia, startups, industry, and government.
- AISI's mandate is to develop indigenous, India-context AI-safety research (grounded in Indian datasets and socio-linguistic diversity) rather than to regulate the AI industry directly.
As the US and China negotiate bilateral AI-safety terms, India's own AISI illustrates the "third-track" approach many middle powers are taking — building sovereign AI-safety capacity and standards rather than being purely dependent on great-power bilateral arrangements.
AI as a National Security and Dual-Use Technology Concern
Frontier AI capabilities are increasingly treated as dual-use technologies with both economic and national-security implications, similar to nuclear or cyber technologies, which is why AI safety talks between major powers are framed in security terms (e.g., autonomous cyberattacks) rather than purely as trade or innovation policy.
Key Details
- Concerns driving major-power AI dialogues include AI-enabled cyberattacks, bio-security risks from AI-assisted design of pathogens, and loss-of-control risks from highly capable autonomous systems.
- Voluntary commitments (rather than binding treaties) have been the dominant governance tool so far — seen in the Bletchley Declaration, the G7 Hiroshima AI Process, and now potentially in the proposed US-China dialogue.
- The lack of a binding verification mechanism for AI safety commitments (unlike, say, IAEA safeguards for nuclear material) remains a structural gap highlighted by security scholars.
The mid-September US-China talks, if held, would be a rare instance of two strategic rivals engaging bilaterally on a dual-use technology risk area, testable as a comparison point against older nuclear/cyber confidence-building measures between major powers.
- Proposed US-China AI safety talks: expected mid-September (2026); agenda and participant list not finalised.
- Bletchley Declaration: adopted 1-2 November 2023 at Bletchley Park, UK; signed by 28 countries plus the EU, including US, China, and India.
- IndiaAI Mission: approved 7 March 2024, outlay ~₹10,372 crore over five years.
- IndiaAI Safety Institute: announced 30 January 2025 under the Safe & Trusted AI pillar of IndiaAI Mission.
- Key discussion themes reported for the US-China talks: AI-enabled cyberattack risk, voluntary self-monitoring by AI labs, and information-sharing frameworks on emerging AI risks.