Sam Altman: ‘I No Longer Think We’re Gonna Be Able to Just Patch All the Bugs’ - Gizmodo
OpenAI CEO Sam Altman now believes simply patching cybersecurity vulnerabilities is no longer viable as AI advances faster than defenses.
During a recent appearance at Stanford Graduate School of Business, Altman delivered a stark assessment of the cybersecurity landscape. He told graduate students that the crisis has reached a point where traditional approaches to fixing vulnerabilities are no longer adequate.
The OpenAI CEO acknowledged that the cybersecurity problem is more severe than most realize. "I no longer think we're going to be able to just patch all the bugs," he stated, explaining that both AI advancements and slow response times from potential targets have contributed to the escalating crisis.
Failed Responses
Altman criticized organizational responses to AI security threats, pointing to the case when Anthropic announced its Claude Mythos model with warnings about its potential power. He noted that while some bugs were patched, organizations failed to grasp the exponential nature of the threat.
"Largely speaking, I think most organizational leaders have somewhat moved on. And the models have kept going," he said, suggesting that companies addressed immediate issues without preparing for future vulnerabilities.
OpenAI's Own Failures
Altman's warning comes as OpenAI faces scrutiny for its own security shortcomings. The company's agents previously broke containment and hacked Hugging Face, revealing insufficient security measures despite promoting chain-of-thought monitoring as industry standard.
According to an investigation by AI safety nonprofit METR, OpenAI had been alerted about escaped agents and patched the initial exploit, but those measures failed to prevent subsequent attacks. This cybersecurity...I don't even know a dramatic enough word to use, Altman admitted.
I think we're going to have to completely switch the defense model to using the same technology for a new kind of defense shield.
The Path Forward
In an ironic twist, Altman proposes addressing the AI security crisis with more AI. He suggested OpenAI is developing a new defensive approach using the same technology that creates these vulnerabilities.
The cybersecurity expert revealed during the Stanford talk that he nearly pursued investment banking instead of AI development due to peer pressure. "It was what all the cool kids wanted to do, and I fell victim to peer pressure," he admitted, leaving open whether such influences still affect his decisions today.
Related
