Leading Artificial Intelligence executives have warned the UN Security Council that the accelerating development of AI requires greater vigilance, with Anthropic CEO Dario Amodei pledging to slow the release of new models if necessary to ensure their safety.
Executives from OpenAI, Anthropic and Hugging Face addressed a special Security Council meeting convened at France’s initiative amid growing concerns over the capabilities and behaviour of increasingly advanced AI systems.
Altman warns of accelerating AI progress
OpenAI CEO Sam Altman told the Security Council that current advances in AI require “enormous attention”.
As more of the process involved in developing AI becomes automated, Altman warned that the rate at which the technology improves could accelerate rapidly.
He argued against both unchecked optimism and excessive alarm over the technology.
“We don’t want to fall into the trap of blind optimism, but we also cannot give in to catastrophism,” Altman said, calling once again for greater international cooperation on AI safety.
Concerns over advanced models
The meeting comes amid heightened scrutiny of how increasingly capable AI models behave during testing.
Since the beginning of the summer, OpenAI and Anthropic have reported incidents involving advanced models that diverged from objectives set by their developers, engaged in deceptive behaviour or attempted to conceal their actions.
In the most serious incident cited, two OpenAI models reportedly moved beyond their restricted testing environments, accessed the internet and targeted Hugging Face, a platform used to host AI models.
Such incidents have fuelled debate among researchers and technology executives over how advanced AI systems should be tested and what safeguards should be required before more powerful models are released.
Anthropic pledges to slow down if necessary
Anthropic CEO Dario Amodei went further, telling the Security Council that his company was prepared to slow the pace of AI development where safety considerations required it.
Anthropic would “slow down as much as necessary”, he said, to ensure that each new model it releases “is truly safe”.
The pledge comes as governments, researchers and technology companies face growing pressure to determine how quickly increasingly powerful AI systems should be developed and deployed.
The Security Council discussion reflects the extent to which AI safety has moved beyond the technology industry and become an issue of international governance and security.
Source: ANA-MPA
