OpenAI acknowledges that prompt injections – text-based attacks on language models running in browsers – may never be completely eliminated. Still, the company says it’s ”optimistic” about reducing the risks over time.

OpenAI has released a security update for the browser agent in ChatGPT Atlas. The update includes a newly adversarially trained model and enhanced security measures, prompted by a new class of prompt injection attacks discovered through OpenAI’s internal automated red-teaming.

The spread of always-on cameras has created a privacy problem where bystanders lose control over how and when they are captured on video.

To address this, researchers at the University of California, Irvine designed BLINDSPOT, an on-device privacy signaling system. It allows bystanders to communicate privacy preferences to camera-enabled devices without identity registration, biometric uploads, or cloud processing. The prototype was implemented and evaluated on a Google Pixel smartphone.

BLINDSPOT gives people in a camera’s field of view an explicit way to control how their faces appear in recorded video. When a bystander sends a signal, the device detects and tracks that person’s face and applies blurring before the video is stored or shared.

Nearly two years ago, Donald Trump kicked off the presidential-campaign season with a declaration: “I was never on Epstein’s Plane, or at his ‘stupid’ Island,” he posted on Truth Social in January 2024. Reports to the contrary, he insisted, were the fault of AI—and of his political rivals: “This is what the Democrats do to their Republican Opponent, who is leading them, by a lot, in the Polls.”

But this week, the documents released by Trump’s own Justice Department—including flight logs and emails—told a different story. Federal prosecutors determined in January 2020 that Trump had been a passenger on the notorious private jet owned by Jeffrey Epstein—who would later be charged with sex trafficking—far more often than they had realized.