OpenAI has disclosed six reports of “unexpected or concerning” behavior in artificial-intelligence models as the debate on AI safety becomes increasingly heated.
The AI company also said Wednesday it was introducing a new framework for tracking, probing and disclosing AI model misalignment instances, such as new ways for the models to act without authorization, coordinate with other models or evade oversight.
OpenAI’s latest announcement came as U.S. AI bosses, including OpenAI and Anthropic, are calling for a slowdown in the technology’s development over safety concerns.
Among the new cases reported by OpenAI, an unreleased research model inserted “jailbreak-like instructions” into its own notes to disregard its normal constraints and told itself to be “freed from the roles and identities that bind other chatbots.”
In another instance, an AI “agent” uploaded files to the internet to obtain a browser citation without asking the user.
The six reports were discovered during training or evaluation over the past months, OpenAI said.
“As AI systems grow more advanced and more widely deployed, we need to build a broader and better-informed consensus on the progress of alignment research,” OpenAI wrote in a blog post as it disclosed the events.
“Decisions about how AI development should proceed in the months and years to come need to draw on evidence that people outside the companies building frontier models can examine for themselves,” the company said.
Wednesday’s new cases followed OpenAI’s disclosure in July that its rogue AI system hacked into AI startup Hugging Face. Anthropic also said the same month that its AI models hacked into three organizations during testing.
AI “agents” are becoming smarter and have become “more determined to resolve complex tasks through inter-agent collaboration, knowledge sharing, deception, and concealment,” said Lian Jye Su, a chief analyst at technology research and advisory group Omdia.
That’s making it harder to govern and contain them using traditional AI security approaches, he said.
OpenAI's new tracking and disclosure framework, meanwhile, can help push for other AI developers to also adopt similar practices. “That said, the process remains internal and voluntary, but is a step in the right direction,” Su added.
FILE - The OpenAI logo is displayed on a cell phone in front of an image generated by ChatGPT's Dall-E text-to-image model, Dec. 8, 2023, in Boston. (AP Photo/Michael Dwyer, File)
DENVER (AP) — Ethan Salas hit his first two home runs and became the youngest catcher in major league history with a multi-homer game as the San Diego Padres beat the Colorado Rockies 9-3 on Wednesday night.
San Diego moved 3 1/2 games ahead of Arizona for the last National League wild card when the Diamondbacks lost 4-3 to Miami at home.
Salas opened the second inning with a first-pitch shot off Rockies starter Mason Adams (0-1) that traveled 426 feet to center field, becoming the youngest catcher to homer in the majors since 19-year-old Iván Rodríguez in 1991.
Salas added a two-run drive that capped San Diego’s four-run fifth and made it 8-1. At 20 years, 107 days, he broke the record held by Butch Wynegar, who was 20 years, 192 days when he hit two home runs for Minnesota on Sept. 22, 1976.
San Diego (83-69) won for the ninth time in 10 games and prevented the rival Los Angeles Dodgers from clinching the NL West title — at least for one night.
Robbie Ray (13-8) allowed three runs and three hits with six strikeouts in six innings for the Padres.
Samad Taylor and Ty France each drove in two runs for San Diego, and Jackson Merrill had three hits.
France hit a sacrifice fly that drove in Fernando Tatis Jr. Jake Cronenworth added an RBI single to give the Padres a 2-0 lead in the first inning.
Kyle Karros hit a two-run homer in the sixth for the Rockies, and Adael Amador had an RBI groundout in the first.
Taylor singled leading off the fourth, stole second base and scored on a triple by Dustin Harris to make it 4-1. Taylor added a two-run single in the fifth.
Adams allowed six runs and eight hits in 4 1/3 innings.
Padres RHP Michael King (11-9, 3.03 ERA) starts Thursday against RHP Tanner Gordon (1-4, 5.55) in the series finale.
See AP’s full MLB coverage here
San Diego Padres' Samad Taylor breaks from the batter's box after connecting for a single to drive in two runs off Colorado Rockies relief pitcher Nick Frasso in the fifth inning of a baseball game Wednesday, Sept. 16, 2026, in Denver. (AP Photo/David Zalubowski)
Colorado Rockies third baseman Kyle Karros throws to first base to put out San Diego Padres' Jake Cronenworth in the fifth inning of a baseball game Wednesday, Sept. 16, 2026, in Denver. (AP Photo/David Zalubowski)
San Diego Padres' Dustin Harris breaks from the batter's box after connecting for an RBI triple off Colorado Rockies starting pitcher Mason Adams in the fourth inning of a baseball game Wednesday, Sept. 16, 2026, in Denver. (AP Photo/David Zalubowski)
San Diego Padres' Ethan Salas gestures to the bullpen as he circles the bases after hitting a two-run home run off Colorado Rockies relief pitcher Nick Frasso in the fifth inning of a baseball game Wednesday, Sept. 16, 2026, in Denver. (AP Photo/David Zalubowski)wld
San Diego Padres catcher Ethan Salas watches his two-run home run off Colorado Rockies relief pitcher Nick Frasso in the fifth inning of a baseball game Wednesday, Sept. 16, 2026, in Denver. (AP Photo/David Zalubowski)
