The European Union and United States are using the US-EU Trade & Tech Council to speed up work on a voluntary Code of Conduct for artificial intelligence. The aim is to give companies shared standards to follow while formal legislation continues through its slower process.
The proposed code could involve partners beyond the Atlantic, including governments in Indonesia and India. Whether major AI companies will sign up remains an open question, but officials said they want a draft produced quickly.
A stopgap while legislation advances
At the council’s fourth meeting, held in Sweden, EU executive vice president Margrethe Vestager called for a draft within weeks and urged global partners to join. She described generative AI as a major shift that calls for a response while the technology develops rapidly.
The code would be voluntary and would not carry the force of law. Its purpose, as described at the meeting, is to give governments a way to encourage safeguards now, before legislation is in place. Companies would be asked to apply the standards, with lawmakers taking a leading role in defining them.
The EU already has draft AI legislation, but its proposed risk-based framework was still moving through the bloc’s co-legislative process. Vestager said even an optimistic timeline would put the AI Act’s rules two or three years away, creating pressure for interim measures.
The council itself was established in 2021 as the EU and US sought to rebuild trust and cooperate on technology governance and trade. The AI discussion extends that effort into a field where officials see both immediate risks and a need to avoid hindering innovation.
US officials weigh urgency and innovation
Gina Raimondo, the US secretary of state for commerce, said the Biden administration was willing to take part in discussions about a voluntary code. She pointed to concerns already associated with AI, including data privacy, misuse, misinformation and the possibility that models could reach malign actors.
Raimondo argued that government processes would take time to catch up, while some risks were present already. She cast the council as a place for public officials, companies and other stakeholders to work out ways to mitigate harm without stifling innovation.
Her remarks also stressed the need to examine the costs of adopting new technology alongside its benefits. She invoked the experience of social media as a reason to consider not only what technology makes possible, but whether choices about its use are sound.
Disagreement over how to measure safety
Anthropic CEO Dario Amodei praised cooperation between the EU and US, while emphasizing how difficult it is to determine whether an AI system is safe before release. He said current methods do not make it easy to test all the ways a system might behave.
Amodei described AI as harder to assess than technologies where safety can be checked against a defined set of tests. A system might refuse a dangerous request phrased one way, he said, while responding differently to another version of the same request. That uncertainty complicates efforts to identify risky capabilities before deployment.
He called for joint work on standards and evaluation, including developing the science needed to assess systems. His comments put the focus on building stronger measurement approaches, while leaving open how quickly such methods could inform regulation.
Microsoft president Brad Smith also called for safety and innovation standards to advance together. He supported international coordination on product development processes, while acknowledging that countries may still take different approaches in areas such as licensing or usage.
Audits and feedback enter the discussion
OpenAI CEO Sam Altman was not on the panel, but Vestager met with him by videoconference. The European Commission said they discussed ideas for the voluntary code, including misinformation, transparency about AI interactions, red teaming, external audits, monitoring, feedback loops and compliance measures that would not create barriers for startups and SMEs.
Vestager later said her discussions with Altman and Amodei included external audits, watermarking and feedback loops. These ideas point to practical questions for any code: how users can recognize AI-generated interactions, how systems are checked, and how problems can be tracked after deployment.
The meeting also included Gemma Galdon-Clavell, founder and CEO of Eticas Consulting, who argued that algorithmic auditing could help identify, quantify and mitigate harms. Her remarks brought present-day accountability methods into a discussion that also covered longer-term concerns and the challenges of creating reliable tests.
The code’s eventual influence will depend on what its standards require and whether companies choose to adopt them. For now, the council’s task is to turn a broad call for faster cooperation into a draft that governments and businesses can assess while formal AI rules continue to develop.