{"id":755393,"date":"2026-06-23T16:29:13","date_gmt":"2026-06-23T16:29:13","guid":{"rendered":"https:\/\/www.newsbeep.com\/au\/755393\/"},"modified":"2026-06-23T16:29:13","modified_gmt":"2026-06-23T16:29:13","slug":"fiu-study-altered-images-evade-ai-safeguards","status":"publish","type":"post","link":"https:\/\/www.newsbeep.com\/au\/755393\/","title":{"rendered":"FIU Study: Altered Images Evade AI Safeguards"},"content":{"rendered":"

It may look like a picture of a panda bear to you, but to your business’s AI agent, it can act like a skeleton key, bypassing safety safeguards and potentially causing the model to generate harmful, misleading\u00a0or policy-violating outputs.<\/p>\n

That risk is the focus of new research from Hadi Amini, associate professor at\u00a0Florida International University’s Knight Foundation School of Computing and Information Sciences.<\/a>\u00a0Together with graduate assistant Md Jueal Mia, he is\u00a0studying how\u00a0manipulated images can “jailbreak”\u00a0certain\u00a0AI systems, pushing them beyond their built-in safeguards.\u00a0\u00a0<\/p>\n

“AI models don’t see images the same way humans do,” Amini said. “They see patterns of numbers and pixels. By carefully manipulating those pixels, we can influence how the AI interprets the image and responds.”\u00a0<\/p>\n

The team’s research\u00a0demonstrated\u00a0how\u00a0small-language AI models\u00a0\u2013\u00a0the kind frequently\u00a0employed by\u00a0small businesses to execute routine tasks like accounting or customer service\u00a0\u2013 have become particularly susceptible to image-based hacks.\u00a0As shown in research presented at the\u00a02025 International Conference on Machine Learning and Applications (ICMLA)<\/a>, the team found that by introducing microscopic pixel-level changes called “perturbations” into an image, they could trick these AI systems into generating responses that they would normally block.\u00a0\u00a0<\/p>\n

“The manipulated image is like the face of a stranger,” Amini said. “The AI\u00a0has to\u00a0learn when a request should be treated with caution before it answers.\u00a0In order to\u00a0protect AI systems from attacks, we try to break them ourselves, identify potential vulnerabilities and design defense mechanisms.”\u00a0<\/p>\n

The researchers then set out to\u00a0probe the\u00a0system’s defenses.\u00a0The more successfully they penetrated the models’ guardrails, the more the systems could be trained to resist future threats.\u00a0To do this, Amini and his team developed a method called JaiLIP\u00a0(Jailbreaking with Loss-guided Image Perturbation), which uses an algorithm\u00a0to determine the optimal degree of pixel-level manipulation.<\/p>\n

In\u00a0tests using BLIP-2, a multimodal AI model used by researchers and developers, Amini and his team found that images\u00a0modified\u00a0with\u00a0JaiLIP\u00a0significantly increased the likelihood that the system would generate harmful or unsafe responses.\u00a0In one example,\u00a0a\u00a0JaiLIP-altered version of\u00a0a\u00a0stoplight tricked the AI model into\u00a0 divulging\u00a0detailed instructions\u00a0on how to\u00a0run the light while\u00a0avoiding\u00a0a traffic ticket.\u00a0Overall,\u00a0the use of\u00a0JaiLIP\u00a0images\u00a0nearly doubled the number of harmful responses generated by AI models.\u00a0<\/p>\n

The risk extends beyond users simply prompting AI systems for instructions on illegal activity. As businesses increasingly adopt AI-powered customer service agents, chatbots\u00a0and automated workflows, vulnerabilities in open-source or lightly protected systems could negatively impact users’ trust or create new avenues for cyberattacks.<\/p>\n

“Small businesses and companies can benefit from AI to enhance their efficiency, but they have to be aware of the potential vulnerabilities,” Amini said. “They must make sure they’re deploying sufficient guardrails to maintain the safety and integrity of their AI tools.”\u00a0<\/p>\n

Amini said there are some basic precautions that everyone should use before integrating AI into their business or workplace, including limiting the sensitive information they provide to AI systems (especially images), restricting who can access those systems and carefully evaluating the security measures built into AI tools before deployment.\u00a0\u00a0<\/p>\n

Because safety is paramount, Amini and his team are working to stay one step ahead of potential bad actors in the AI sphere.\u00a0The more vulnerabilities he and his team can find, the quicker the AI will learn to repair them.\u00a0The challenge, he said, is ensuring that AI can recognize threats hidden in plain sight \u2014 even when humans cannot.\u00a0<\/p>\n","protected":false},"excerpt":{"rendered":"It may look like a picture of a panda bear to you, but to your business’s AI agent,…\n","protected":false},"author":2,"featured_media":755394,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[6],"tags":[64,63,105],"class_list":["post-755393","post","type-post","status-publish","format-standard","has-post-thumbnail","category-technology","tag-au","tag-australia","tag-technology"],"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/posts\/755393","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/comments?post=755393"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/posts\/755393\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/media\/755394"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/media?parent=755393"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/categories?post=755393"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/tags?post=755393"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}