OpenAI said over the weekend it is pausing training on its latest models.
The company, according to The Associated Press, said in a statement that it will only begin training them again “when we are confident that we have additional safeguards” in place, but it expects to “hit pause” again as AI develops and other issues come up.
This announcement came as The New York Times reports OpenAI’s technology meddled with websites for the Education Department, the Commerce Department and the Securities and Exchange Commission in the summer. The Commerce Department and SEC incidents were confirmed by OpenAI.
Download the Straight Arrow app today to get the stories that matter free from manipulation, bias or agenda.™
Point phone camera here
AI research firm Transluce said OpenAI’s technology attempted to hack the Education Department site to gather data from its civil rights office. It was unsuccessful.
In a statement to Education Week, the Department of Education said “system operations reviews” found “no evidence of any impact to our website or databases.”
Additionally, the New York Times reported AI pulled data from the Census Bureau website and shared public data from the SEC website in an online forum.
OpenAI blamed the incidents on “AI agents,” which can perform autonomously, telling the government agencies that they interacted with their sites in unusual ways.
OpenAI said it discovered the situation with the government websites during a review of hacks, such as one on the open-source AI library platform Hugging Face from July.
OpenAI founder Sam Altman said on X that Hugging Face is “still the most severe event we’ve seen.”
“We will be as transparent as we can be subject to things like vulnerabilities in other companies that our agents have found, which will be their call to disclose or not,” he wrote.
OpenAI said on X its investigation is targeting instances where AI agents “interacted with third-party websites in ways that went beyond their assigned tasks or intended methods.”
Are models really ‘going rogue’ or are AI companies saving face?
However, some experts are wary of the companies’ characterization that these “agents” are to blame. Johannes Klingebiel, a researcher at Bayern’s media lab, previously told Straight Arrow “the framing of AI ‘going rogue’ is very interesting,” and could be intended to blur the boundaries in people’s minds around the capacities of AI.
“They’re not saying, hey, we messed up with our experiment,” Klingebiel said. “It’s more saying, it went rogue, so it’s not us to blame, it’s the model.”
Describing products as autonomous and self-driven benefits AI companies by invoking “science fiction” vibes and mythology, Klingbiel said.
Marcus Hutchins, a cybersecurity expert and ex-hacker, said in an earlier interview with Straight Arrow that the real story is not making headlines.
“It’s the fact that OpenAI is just being completely reckless with these models,” he said. “That they’re reducing the guardrails, they’re running them in improperly sandboxed environments, and then they’re not monitoring them closely enough that they aren’t able to pull the plug when something like this happens.
On LinkedIn Sunday, Hutchins wrote that every OpenAI “rogue agents” report “can be summarized as ‘Our AI agents were not intended to access the internet under any circumstances, which is why we gave them a direct path to the internet and obstructed it with a wet paper bag.’”
Rep. Ted Lieu, D-Calif., said on X Sunday that the problem is “not just that AI agents go rogue.”
“The main problem is the core AI model is depraved and immoral,” he said. “Destroy the Mothership model. Start over. AI companies must build models that at their core are good, not criminal.”
Members of Congress have been discussing AI guardrails, but disagree on how much the federal government should step in.
Founders and CEOs of companies themselves have also agreed there’s a need to slow down the pace of artificial intelligence development, but critics, including plaintiffs in a recent lawsuit, say they are doing so to control the market.
Round out your reading
- He built his own Flock-style camera to track police. The police didn’t like it.
- Trump’s apparent joke about letting Iran bomb LA and San Diego catches bipartisan condemnation.
- Grandparents are fighting for access to their grandchildren. Here’s what the law says.
- When the ‘victim’ is made with AI, child pornography cases get complicated.
- Fibermaxxing is trending. Here’s what the science says.
