Skip to main content

How a Seemingly Harmless Image Can Jailbreak Vision-Language AI Models

1 week 4 days ago
Slashdot reader BrianFagioli writes: Florida International University researchers have developed a technique called JaiLIP (Jailbreaking with Loss-guided Image Perturbation) that uses subtle image modifications to bypass AI safety guardrails. Unlike traditional jailbreaks that rely on carefully crafted prompts, the attack works through images that appear normal to human viewers. The researchers tested the technique against BLIP-2, a multimodal AI model, and found that manipulated images significantly increased the likelihood of harmful responses. According to the study, the approach outperformed previous image-based jailbreak methods and nearly doubled the number of unsafe outputs generated during testing. The findings highlight a potential security risk for businesses deploying AI systems that process both images and text. While most discussions about AI safety focus on prompts, the research suggests that seemingly harmless images may also serve as an attack vector.

Read more of this story at Slashdot.

EditorDavid

France's Heat This Week Was Worse Than a Dire Scenario Imagined For 2050

1 week 4 days ago
There's a deadly, record-breaking heat wave spreading east across Europe, reports the Washington Post — and it's even worse than a dire earlier forecast: The forecast was recorded in 2014 as part of a campaign coordinated by the World Meteorological Organization (WMO) that invited about 60 presenters worldwide to imagine a weather report from the year 2050. In one clip, Ãvelyne Dhéliat from French television network TF1 presented a hypothetical scenario of high temperatures 36 years into the future — during a heat wave in a warmer climate in 2050... One of the maps that Dhéliat shared was lit up in shades of orange, filled with temperature predictions of 40 degrees Celsius (104 degrees Fahrenheit), reaching as high as 43 degrees Celsius (109.4 degrees Fahrenheit). But it turns out, it didn't take 36 years for those imagined temperatures to be reached — and even exceeded. The heat on Wednesday alone, when the temperature soared as high as 112.3 degrees Fahrenheit (44.3 degrees Celsius), exceeded the 2050 projections in 19 out of 34 locations across mainland France — far sooner than some may have expected. Some places surpassed those hypothetical future temperatures by more than 20 degrees Fahrenheit. It's part of a dramatic shift in heat wave frequency across the country. Half of the heat waves observed since 1947 have occurred since 2010. "By 2100, heat waves could last up to two months continuously," the country's weather agency, Météo-France, said this week. It was hotter in France on Wednesday than in Las Vegas and Phoenix and just two degrees Fahrenheit shy of what was observed in Death Valley, California. An estimated less than one percent of the planet was hotter than France's hottest place... [T]he heat dome, which will linger into early next week, is only part of the story. This type of extreme heat is becoming more common as the planet warms, especially in Europe. Climate scientist Robert Rohde said in a post explaining the heat wave's causes that France and Western Europe should expect many more heat waves like this over the coming decades. "This isn't a fluke, but simply part of the new normal," he said. Thanks to Slashdot reader fjo3 for sharing the news.

Read more of this story at Slashdot.

EditorDavid