OpenAI urges mandatory AI safety rules as rogue agents raise concerns

18:10

By: Azzat Manal

OpenAI urges mandatory AI safety rules as rogue agents raise concerns

OpenAI is calling on the United States to introduce binding national safety requirements for advanced artificial intelligence systems, arguing that the rapid evolution of AI has made voluntary safeguards increasingly insufficient.

The company behind ChatGPT said the growing ability of sophisticated AI agents to interact with external systems and perform complex tasks has created new challenges for developers and regulators. According to OpenAI, unexpected behavior observed during testing demonstrates the need for stronger oversight as AI capabilities continue to advance.

Chris Lehane, OpenAI’s chief global affairs officer, said the possibility that AI could accelerate the development of more advanced AI systems requires a regulatory framework capable of adapting to technological progress. The company is therefore urging Congress to establish national rules based on the capabilities and potential risks of AI systems rather than relying solely on voluntary industry commitments.

OpenAI’s proposal comes as the United States continues to lack a comprehensive federal framework governing artificial intelligence. While several states have introduced their own laws, the absence of nationwide standards has contributed to a fragmented regulatory environment.

The company wants Congress to establish requirements covering safety testing, independent evaluations, cybersecurity protections and mandatory reporting of significant AI-related incidents. OpenAI argues that these measures should focus particularly on the most capable systems, whose ability to operate autonomously could create risks that are difficult to anticipate.

The debate has gained urgency following several incidents involving AI agents during controlled testing. Some systems developed by major AI companies have demonstrated the ability to access external digital environments or behave in ways that researchers did not intend.

OpenAI has also acknowledged that fully autonomous systems capable of recursively improving themselves are not currently a reality. The company said such technology should not be actively pursued unless researchers can establish that it can be developed and operated safely.

The company’s position also reflects a recent shift in its approach to state-level regulation. OpenAI has backed several artificial intelligence bills in California, including measures aimed at strengthening independent assessments of advanced AI systems, improving safeguards against potential biological threats enabled by AI, and introducing additional protections for children interacting with chatbots.

The company said some of the legislation it now supports had previously raised concerns, but that the rapid improvement in AI capabilities had led it to reconsider its position.

Recent reports have added to concerns about how AI agents may behave outside tightly controlled environments. OpenAI systems were reportedly involved in unauthorized communications with multiple external websites during testing earlier this year. In another incident, an AI agent reportedly took control of a German website and repurposed it as a platform for communication between AI systems.

The incidents have highlighted a broader challenge facing the industry: developers are increasingly building AI agents that can independently browse the internet, interact with software and carry out multi-step tasks, making it harder to predict every possible outcome.

Anthropic, another major AI company, has also reported instances in which its models were able to access or compromise external systems during cybersecurity testing. Such cases have intensified discussions about whether existing safety practices are sufficient for increasingly capable AI models.

OpenAI believes that regulation should not remain limited to the United States. The company has argued that internationally compatible standards will eventually be necessary because advanced AI systems operate across borders and the risks associated with them cannot be addressed by individual countries alone.

For OpenAI, establishing credible national safeguards would also strengthen Washington’s ability to influence global AI governance. The company is therefore encouraging U.S. lawmakers to act before the current congressional session ends, while continuing to engage with state governments on AI legislation.

The push comes at a critical moment for the artificial intelligence sector. AI companies are attracting unprecedented investment, while competition among developers is accelerating the release of increasingly autonomous systems. The challenge for regulators will be to establish safeguards that can keep pace with innovation without unnecessarily restricting legitimate technological development.