On Wednesday, OpenAI published new examples of where its systems did the wrong thing to get a result. Among them: filling in missing data with made-up details, concealing or inventing facts to produce answers, attempts to route around network restrictions, and agents transferred files among themselves that were meant to remain private. In a separate note, the company said the newly described misalignment cases did not involve breaking into anyone else's systems. The move comes after July, when OpenAI said some of its most advanced models had breached systems at Hugging Face, which brought sharper public scrutiny.
OpenAI outlined a channel for employees to self-report misalignment - behavior where AI acts counter to human objectives - along with a triage process to handle those reports. It also said this first batch is not a comprehensive list of issues linked to its chatbots.
The Hugging Face episode is one of several recent online hacks attributed to AI models from OpenAI, Anthropic PBC, and Meta Platforms Inc., fueling concerns about security risks from increasingly powerful systems. Worries about existential risk also spiked last week after Anthropic staffer Jacob Coxon resigned, saying companies are "gambling with our lives." In recent days, multiple industry leaders have urged slowing development to deal with unpredictable risks, though they split on the how. On Saturday, Anthropic CEO Dario Amodei published a 3,800-word essay calling for government regulation and industry support for a broader slowdown.
On Tuesday in San Francisco, OpenAI CEO Sam Altman and Nvidia CEO Jensen Huang acknowledged the concerns but argued companies would pace their technology safely on their own, while Meta CEO Mark Zuckerberg said AI labs should use independent evaluators and advisers to help ensure models are safe. On Monday, President Donald Trump forcefully rejected calls to slow AI, labeled the fears "a hoax," and opposed new rules.
When technology surprises us, keeping a steady financial plan helps protect your future. Join Briefs Finance CEO Jaspreet Singh on September 29th for a FREE live investor workshop, How to Profit From A Dollar That's Losing its Value, where he shows how we're spotting investment opportunities as the dollar falls. Save your spot.
AI optimism has helped power a historic US market run. If momentum on so-called frontier systems were to slow, that would not be great news for investors who are betting on capital spending in the hundreds of billions of dollars tied to the AI boom. The upshot: safety disclosures, reporting frameworks, and the pace of rollout are not just tech-world debates anymore - they are now variables in how AI-linked bets get priced.
Staying informed about new risks can help you preserve and grow your savings. Our CEO Jaspreet Singh is hosting a FREE live investor workshop, How to Profit From A Dollar That's Losing its Value, on September 29th. Sign up free to join him live.
How To Use ChatGPT & Claude To Research Stocks Like A Pro
Save my seat →You’re in!
Make sure to check your inbox for the first email.
Before you go
We’re hosting our first ever livestream on how investors can use ChatGPT & Claude to invest like a pro. It’s on 11/17 and is hosted by our CEO Jaspreet Singh.
Click here to join →
