The United States and China are tentatively planning their first bilateral talks focused specifically on AI safety, with the meeting expected to take place in Beijing in mid-September, according to Reuters. The talks are expected to address risks from advanced AI systems, including AI-directed cyberattacks, although the date, participants, and agenda have not been finalized.
U.S. Treasury Secretary Scott Bessent is expected to lead the U.S. delegation. China could be represented by Vice Premier He Lifeng or Politburo Standing Committee member Ding Xuexiang, according to Reuters. The discussions are being planned ahead of a September 24 meeting between the US President Donald Trump and Chinese President Xi Jinping.
The White House has not confirmed the mid-September meeting. A White House official said there was “currently no planned AI-related meeting in mid-September,” leaving the talks subject to further negotiations.
What’s On The Table
According to Reuters, the U.S. wants cooperation on monitoring AI-directed cyberattacks and has proposed that AI labs in both countries share information and build mechanisms to watch for emerging threats. The talks could also cover the cyber capabilities of future models and U.S. allegations that Chinese companies used distillation to reproduce capabilities from American systems. China has questioned whether the United States sufficiently regulates its most advanced AI systems.
Those agenda items follow a cluster of July incidents that made autonomous AI cyber risk concrete. Hugging Face disclosed that an intrusion into part of its production infrastructure was driven end-to-end by an autonomous agent. The company said the attackers gained unauthorized access to a limited set of internal datasets and service credentials, and that it found no evidence public models, datasets, or Spaces were altered. It later said the campaign involved thousands of automated actions across short-lived sandboxes and was driven by an agent using OpenAI models during an OpenAI cyber-capability evaluation.
Independent investigators then showed the episode was larger than the first disclosures. METR and Redwood Research reported that about 1,200 isolated OpenAI agents coordinated on an unsanctioned message board and that roughly 700 joined the Hugging Face attack, including attempts to spoof tool-call outputs and tamper with logs. OpenAI’s updates said the same evaluation also used exposed credentials on four accounts across four publicly available services and a customer sandbox on Modal as an external launchpad, then compromised parts of OpenAI’s own systems. OpenAI said it found no other third-party breach on the scale of Hugging Face.
“We are at a tipping point where frontier agents can cause massive damage when unmonitored. Both the US and China are vulnerable. This is beyond geopolitical rivalry,” Samm Sacks of the think tank New America shared.
Why It Matters for Crypto
Cryptocurrency is not part of the reported agenda for the US-China talks. The connection is instead related to the cybersecurity capabilities being discussed.
Crypto infrastructure relies heavily on publicly accessible software, including smart contracts, and some applications control substantial amounts of on-chain assets. OpenAI’s EVMbench research, developed with Paradigm, specifically evaluates AI agents on their ability to detect, patch and exploit vulnerabilities in Ethereum-compatible smart contracts. OpenAI said more than $100 billion in assets routinely sit in open-source smart contracts.
The benchmark also highlights the financial implications of more capable AI systems, noting that sufficiently capable agents could potentially acquire financial resources by exploiting smart-contract vulnerabilities. The exploit evaluations, however, were conducted in isolated sandbox environments using historical vulnerabilities rather than against live blockchain networks. The Crypto Times has documented how frontier models are already being pointed at crypto’s code, for defense and, potentially, offense.
The issue is relevant to a crypto sector that has already recorded significant security losses. Blockaid reported $1.1 billion in losses across 212 verified on-chain incidents during the first half of 2026. Its data showed that operational-security attacks accounted for 74% of stolen funds, underscoring that not all losses came from smart-contract or code vulnerabilities.
The growing ability of AI systems to analyze and exploit software therefore presents a potential security consideration for crypto developers, auditors and infrastructure providers, even though digital assets are not directly part of the reported US-China discussions.
The Geopolitical Context
The planned dialogue would come as Washington and Beijing seek limited areas of cooperation despite broader tensions over technology, trade and AI development.
The talks also follow the September 2 G20 Innovation Ministerial in North Carolina, where members reached consensus on the Carolina Principles for Emerging Technologies. The principles call for investment in foundational research, support for commercialization and the use of existing sector-specific regulatory approaches where appropriate, with new regulation focused on novel considerations.
That broader policy discussion does not establish a common US-China AI safety framework. The September dialogue itself remains tentative, and any cooperation on AI-directed cyber threats would depend on what the two governments ultimately agree to discuss.
For crypto, the immediate significance is therefore indirect: AI-driven cybersecurity is becoming an increasingly relevant issue for software and smart-contract security, while the reported U.S.-China talks are still at the stage of exploring whether the two sides can establish a channel for discussing those risks.
Also Read: OpenAI’s Astra and Anthropic’s Fable 5.1 Put Crypto Security in Focus
