Here’s an uncomfortable question for the generative AI era: if an AI creates an image, can anyone actually point to the specific images that influenced it? A new study from MIT’s Computer Science and Artificial Intelligence Laboratory suggests that, for sufficiently large models, the answer may often be no. Researchers Dai and Gifford describe the phenomenon as “attribution decay” in their open-access paper, published today in Nature Communications. It refers to how the influence of any individual piece of training data becomes increasingly difficult to detect as the size of the dataset grows.
The researchers wanted to test something more precise than simply asking whether an AI model had been trained on a particular image. Their approach was essentially a counterfactual: what happens if a specific piece of training data is removed?
They found that, as models and datasets scale, removing an individual image can make little or no measurable difference to the generated result. The same can apply when removing an entire artist’s work or images of a particular person. In other words, the model may have learned broad visual patterns from an enormous pool of data without any single image being clearly responsible for a particular output. MIT researchers Zheng Dai and Professor David Gifford argue that if removing a particular piece of data doesn’t change the output, it becomes difficult to attribute that output to the removed example meaningfully.
To be fair, the study doesn’t mean training data is irrelevant, nor does it settle the broader debate over whether AI companies can use copyrighted material without permission. Instead, its focus is narrower: whether a specific training image can be shown to have directly influenced a specific AI-generated output. A model may not reproduce any particular artist’s work while still relying on millions of images to learn things such as composition, lighting, textures, and artistic styles.
The more interesting takeaway is that this connection becomes increasingly difficult to trace as datasets grow. An AI-generated image may draw on patterns learned from an enormous pool of material without having a clear, identifiable source image behind it. The model may have learned from everyone, while leaving no single image with an obvious fingerprint on the final result.
AI companies spend enormous amounts of time convincing us their technology is safe and secure. But what happens when those same companies become targets of increasingly sophisticated AI-powered attacks? Anthropic, the company behind Claude, is now facing that question. According to a Washington Post report, cybersecurity startup Cathedral discovered vulnerabilities that let its researchers access sensitive information in Anthropic's internal systems.
The researchers used AI agents to identify potential entry points in less than 12 hours. Interestingly, they didn't need to crack Claude itself. Instead, they exploited weaknesses in third-party business applications that Anthropic uses for everyday operations. Anthropic's security testing program disclosed the findings, which allows outside researchers to identify and report vulnerabilities.
Military working dogs charge into places the rest of us would flee. So, I understand why anyone would want to breed tougher, healthier dogs for the job.
Warhound, a new startup from Colossal Biosciences co-founder Ben Lamm and Anduril co-founder Palmer Luckey, has stepped out of stealth with $200 million to clone and gene-edit working dogs for American troops, police departments, and emergency services (via TechCrunch).
AI can describe a picture, identify objects, and answer questions about what a camera sees. But when a blind user needs a quick answer, getting an entire spoken essay about the scene isn't exactly helpful. Researchers at the University of Maryland (UMD) have found that AI-powered visual assistance can leave blind users doing too much work to get the information they actually need.
The study, titled Say It My Way: Exploring Control in Conversational Visual Question Answering with Blind Users, examined how people interact with Be My AI, an AI-powered feature of the Be My Eyes app. The researchers found that users often received lengthy descriptions that didn't adequately address their immediate goals, forcing them to ask follow-up questions or find ways to steer the AI towards a more useful answer.