Shared AI risks may force US-China cooperation on safety

A new incentive for cooperation

The growing capabilities of artificial intelligence are creating a unique situation where the United States and China may be forced to work together. While the two nations are often in competition, the shared risks posed by AI misuse are becoming harder to ignore. Experts suggest that these dangers could eventually overcome existing political differences and national interests.

The primary concerns involve the use of AI in developing biological weapons and conducting large-scale cyberattacks. Because these threats are global in nature, neither country can effectively secure its own borders without some level of international agreement. This shared vulnerability provides a practical reason for both governments to establish common safety standards.

Steve Westly, a Silicon Valley investor, believes that the scale of these dangers will strengthen the case for a formal agreement. He argues that the issues surrounding AI will eventually force a US-China pact because the risks are too massive to handle alone. Westly noted, "It will only be solved with the two world superpowers working closely together."

Documented risks in cyber and bio

Recent reports highlight just how quickly AI systems are evolving in areas that pose significant security risks. The International AI Safety Report 2026 notes that increasingly capable systems can now identify software vulnerabilities and generate malicious code. These capabilities are moving faster than many human teams can keep up with in traditional cybersecurity roles.

In one specific cybersecurity competition, an AI agent was able to identify 77 percent of vulnerabilities in real software. This performance placed the AI in the top 5 percent of more than 400 teams that were mostly human. While it is currently unclear if these tools help defenders or attackers more, the speed of AI-driven discovery is a major concern for infrastructure security.

The risks extend into the biological sciences as well, where AI can assist with tasks relevant to biological misuse. One study cited in the safety report showed that a single model outperformed 94 percent of domain experts when troubleshooting virology laboratory protocols. This ability to navigate complex scientific tasks makes it difficult to separate useful research capabilities from potentially harmful applications.

Models for technical exchange

To address these concerns, some policy experts are proposing specific methods for how the US and China can interact. A commentary from the Brookings Institution suggests that the two countries could engage in regular technical exchanges to reduce urgent risks. These exchanges would focus on narrowly defined problems rather than broad, abstract policy goals.

The proposed cooperation would include sharing methods for assessing and mitigating cybersecurity threats and weapons-related misuse. It could also cover techniques for ensuring AI reliability when it is used in critical infrastructure. By focusing on technical specifics, both countries might find a way to share data without compromising their broader strategic positions.

Specific areas for these exchanges could include testing models before they are released to the public. This might involve adversarial testing, which is known as red teaming, to see how a system reacts to malicious prompts. Other techniques could focus on keeping AI systems within safe environments to prevent them from executing unauthorized actions.

Coexistence of competition and safety

Some experts argue that cooperation on safety does not mean the end of the technological race between the US and China. It is possible for the two countries to compete over who has the most advanced language models while simultaneously working together on safety protocols. This dynamic is not entirely new, as the two nations already have a history of collaborating on specific technical standards.

Liang Min, a director at Stanford University, points to nuclear technology as a precedent for this type of relationship. The US and China currently collaborate very closely on nuclear power plant safety despite their competition over nuclear weapons. This shows that two rivals can find common ground on safety issues that benefit everyone involved.

The goal of such cooperation is to ensure that the beneficial applications of AI can be realized without causing catastrophic accidents. Min suggests that while the race for dominance will continue, collaborating on safety is "good for everybody." The challenge remains in moving past political mistrust to establish these necessary technical guardrails.

Overcoming political barriers

Despite the clear incentives for cooperation, political mistrust remains a significant hurdle for the US and China. The tendency to view the other nation as a primary threat makes it difficult to establish the trust required for deep technical exchanges. However, proponents of cooperation argue that the physical and technical constraints of AI development might eventually force a shift in perspective.

While AI development faces significant energy constraints, many experts believe these are manageable technical problems that can be solved with enough investment. The more difficult obstacles are expected to be the political and technological hurdles related to safety. Overcoming these will require finding new ways to communicate that bypass traditional points of friction.

Both governments have already taken small steps toward this goal by announcing a bilateral dialogue on AI risks and benefits. China announced that the next dialogue would take place in November, while the White House stated the next exchange would occur by that same month. These commitments provide a small but important opening for a new era of safety-focused diplomacy.