Hacker stole secrets from OpenAI

On July 4, 2024, the New York Times revealed that OpenAI had suffered an unknown breach in early 2023.

The hacker reportedly did not gain access to the company's systems, but stole discussions from an employee forum. OpenAI did not publicly disclose the incident or notify the FBI, as it claims no customer or partner information was stolen and the breach was not considered a threat to national security.

Read also: ChatGPT-4o led users to scam site

The company concluded that the attack came from an individual not affiliated with any foreign government.

However, the incident sparked internal discussions among staff about how seriously OpenAI takes security concerns.

“After the breach, Leopold Aschenbrenner, OpenAI’s technical program director, focused on ensuring that future AI technologies would not cause serious harm. In a memo to OpenAI’s board, he argued that the company had not done enough to prevent the Chinese government and other adversaries from stealing its secrets,” the NYT reports.

Earlier this year, he was fired, ostensibly for leaking information, though more likely because of the memo. Aschenbrenner has a slightly different take on the official story of the leaks. In a podcast with Dwarkesh Patel (June 4, 2024), he said: “OpenAI has told employees that I was fired because of a data leak. I and others have asked them to clarify what that leak was. Their response was: Sometime last year, I had written a brainstorming paper about readiness, security, and the necessary steps for the path to AGI. I shared it with three external researchers for feedback. This action was considered a leak. Before sharing it, I reviewed it for any sensitive content. The internal version contained a reference to a future cluster, which I removed from the external copy.”

See more: ESET Report: Infostealers use AI to extort money

Clearly, OpenAI is not a monotonous entity, as it encompasses many different views on how it should operate, its ideal function, and its direction. The concern is not so much focused on OpenGPT (which is generative artificial intelligence) as on the future of AGI (artificial general intelligence).

The former transforms information acquired (usually from the internet), while the latter has the capacity for original thinking. Although Gen-AI (general AI) is not considered a threat to national security, it could increase the scale and sophistication of existing cyber attacks.

AGI is a completely different matter. It will have the potential to create new threats in cyberspace, the battlefield, and intelligence. Leading AI companies and technologies, such as OpenAI, DeepMind , and Anthropic, are racing to be first to market. The concern about the breach in 2023 is that it could signal a lack of security preparedness that could jeopardize national security in the future.

“A lot of the drama comes from OpenAI’s belief that they are actually creating AGI. This is not just a marketing claim,” Aschenbrenner said. “What confuses people is the contrast between believing in AGI and not seriously considering some of the consequences. This technology is going to be incredibly powerful, for both good and bad, which is a national security issue. Are your secrets protected by the CCP? Does America control the core AGI infrastructure, or is it in the hands of a Middle Eastern dictator?”

Read also: ChatGPT Mac App: Saves user conversations in plain text

As we approach the development of Artificial General Intelligence (AGI), cyber threats will shift from criminals to elite nation-state hackers. Yet we repeatedly see that our security is not adequate to deal with them. After a relatively small breach at OpenAI, Aschenbrenner expressed sincere and substantive security concerns – and for that, it seems, he was fired.

Source: securityweek