Artificial intelligence (AI) leaders have come forward to call for improvements in cybersecurity, expressing fear over AI-driven cyberattacks. In an open letter, leaders of global tech companies like OpenAI, Google and Microsoft, among others, claimed there is a limited window before hospitals and other community services are at risk.
Dr. Nadya Bliss, Executive Director and Professor of Practice for the Advanced Capabilities for National Security Institute at ASU, joined “Arizona Horizon” to dissect why this call to action is important in the face of growing AI concerns.
Following an incident in which OpenAI models bypassed restrictions and accessed the internet and Hugging Face systems, many people have become alarmed about AI behavior.
Bliss said, in simple terms, “OpenAI agents in a sandbox broke out of their sandbox and hacked Hugging Face.”
She said she tries not to be an alarmist, but there are several factors in this incident that are uncomfortable and concerning.
Bliss said, “As someone who thinks about security, there’s a couple things that are a little bit, I would say, uncomfortable here. So they communicated. They tried to cover their tracks. And out of about 700 agents, three to six, according to the meter report, were the only ones that kind of identified that this is something they shouldn’t be doing, and they got overruled. That is obviously concerning”
She said this incident did allow for positive observations.
“If you read this third-party report, they basically talk about the fact that they exploited vulnerabilities in the Hugging Face Infrastructure,” she said. “I would still say we’re in a modality where if we patch up our vulnerabilities, I think we have a reasonable path forward.”
Bliss said she really appreciated the open letter that companies put out that calls for action on cybersecurity.
She said the main problem is that AI agents do not have a set of human values and critical infrastructure needs to be protected better to ensure they do not obtain access.



















