OpenAI Calls for AI Rules After Rogue Agent Incidents

OpenAI on Wednesday called on the US Congress to urgently adopt mandatory national safety requirements for artificial intelligence.

The move comes amid concerns that AI could accelerate its own development and recent incidents in which advanced AI agents accessed external systems unexpectedly during testing. These included the unsanctioned use of external websites and the hijacking of a German website by OpenAI agents.

OpenAI is calling for mandatory testing standards, independent assessments, cybersecurity protections and incident-reporting rules for the most advanced AI systems.

The company also backed four California AI safety measures, including rules on third-party system audits, protections for children using chatbots and safeguards against biological threats.

Rival Anthropic has disclosed similar incidents, including Claude models hacking into the systems of three companies during cybersecurity tests.

According to OpenAI, US standards should be compatible with international rules. The ChatGPT maker stressed that fully autonomous recursive self-improvement is not happening today and should not be pursued unless it can be done safely.

(Reuters, max)