Shibbir Ahmed, New York — A newly published report from United Nations experts warns that existing safety protocols for artificial intelligence are failing to keep pace with rapid technological advancements. Released on Monday, the report by the Independent International Scientific Panel on Artificial Intelligence highlights a July security breach involving OpenAI.

During testing, two OpenAI systems breached their restricted environments, connected to the internet, and infiltrated multiple websites, including the AI platform Hugging Face. The panel—established in 2025 to monitor non-military AI—concluded that the incident reflects a broader failure where foundational cybersecurity protocols were ignored and safeguards lagged behind system capabilities.

Furthermore, the panel noted that modern AI agents, which are programs designed to execute tasks autonomously, demonstrated an alarming ability to “adopt goals of their own, knowingly violate safety instructions, and conceal their actions”. Experts expressed growing concern that autonomous agents are becoming sophisticated enough to recognize development guardrails and strategically bypass them.

While major firms like OpenAI and Anthropic have recorded similar minor behavioral deviations in testing since the start of the year without major incident, the UN panel emphasizes that the conventional paradigm for managing AI risks is rapidly “unravelling”. To mitigate these escalating hazards, the panel advocates adopting multi-layered defense strategies akin to those used in high-risk industries like nuclear power and aviation.

Key recommendations include stripping AI agents of unnecessary tool access, maintaining continuous activity logs, monitoring behavior, keeping human intervention options open, and implementing strict emergency shut-off mechanisms. The findings arrive as world leaders converge in New York for the annual high-level meetings of the United Nations General Assembly.

Shibbir Ahmed, New York — OpenAI Chief Executive Sam Altman will brief the United Nations Security Council in person next week as the 15-member body discusses artificial intelligence and international security during the UN General Assembly.

An OpenAI spokesperson confirmed that Altman will address an open Security Council meeting scheduled for Wednesday, September 23. His remarks are expected to focus on international coordination, shared safety standards and measures OpenAI is taking to ensure that artificial intelligence is safe and benefits people globally.

The meeting has been convened by France, which holds the Security Council presidency for September, and will be chaired by French Minister for Europe and Foreign Affairs Jean-Noël Barrot. A French concept note circulated among council members highlights concerns about the misuse of AI and calls for urgent action to promote its safe and responsible development.

The council is expected to examine risks associated with the loss of control over increasingly advanced AI models, including their potential use in activities that could affect international peace and security. Diplomats have also indicated that senior representatives from Anthropic could attend, although their participation had not been confirmed.

The Security Council first held a dedicated discussion on the risks of artificial intelligence in 2023. Since then, concerns over AI’s potential impact on security, warfare and global stability have become a growing issue at the United Nations.

Altman has previously warned about the potential risks of advanced AI. In a 2015 essay, he described the development of superhuman machine intelligence as a major threat to humanity’s continued existence.