Tufts publishes register pairing more than 100 AI harms with available remedies
Developed with Oxfam, the work-in-progress resource assesses severity, reach and timing. Its creator argues that existing laws matter only if people can use them.
Loading page…
Developed with Oxfam, the work-in-progress resource assesses severity, reach and timing. Its creator argues that existing laws matter only if people can use them.
Listen to this story
The register’s central policy question is whether people harmed by AI have a practical remedy—not just how severe a risk might be. It records 103 incidents and weighs each by severity, reach and timing, then considers available defenses, compensation and penalties. The framework highlights a gap between laws that may already apply and the capacity to identify and enforce them. Tufts plans monthly updates, tracking both newly documented harms and remedies still needing development.
The register includes harms ranging from wrongful arrests and data-center water use to rogue AI agents hacking systems.
Tufts Now describes uncontrollable AGI as the highest-severity risk and the one with the widest gap between potential harm and response readiness.
Bhaskar Chakravorti argues that severe risks with existing remedies warrant faster action, while institutions should prepare for serious risks without remedies.
Tufts University’s Digital Planet initiative has made public a register that pairs more than 100 AI-related harms with a practical question: is there a remedy? Developed with Oxfam, the work-in-progress Register of AI Risks & Remedies assesses potential damage alongside the defenses available—not simply which threats sound most alarming.
Bhaskar Chakravorti, dean of global business at Tufts’ Fletcher School, led the project with a team at Digital Planet, assisted by AI. In a November 16 interview with Tufts Now, he said the team put the unfinished register into public view to bring attention to the full range of risks, including immediate harms that can be overshadowed by warnings about human extinction.
The register draws on regulatory actions, court proceedings, media coverage and technical reports. Its entries span wrongful arrests, water use by data centers and rogue AI agents hacking into systems. Chakravorti said the team had captured 103 risk incidents, giving the project a scope that reaches beyond failures inside AI software to consequences for people and resources.
Rather than treating every entry as an equivalent warning, the register scores each harm along three dimensions. It then examines whether a remedy, such as a law, is available to address it.
They may seem more mundane, but they’re not mundane if you’re the individual on the wrong side of that risk.
Bhaskar Chakravorti, speaking to Tufts Now
Tufts Now identifies uncontrollable artificial general intelligence as the register’s most severe risk with the widest gap between potential harm and readiness to respond. Asked whether that should dominate efforts, Chakravorti argued for reserving substantial attention and resources for other risks that are more immediate and difficult to address.
He illustrated the stakes with a hypothetical patient pushed down a hospital queue because an algorithm relied on data biased against their ethnicity, socioeconomic class or race. His argument was not to dismiss existential danger, but to keep individual harms visible alongside it.
For policymakers with limited resources, Chakravorti proposes weighing severity against the tools already available. Severe risks with ready remedies should receive faster action, he said. Where serious risks lack remedies, institutions need to be built before harm occurs. His framework also asks whether victims can receive compensation and whether those deploying AI without necessary safeguards can be penalized.
In most of the incidents captured, Chakravorti said, an applicable law already exists. But using it requires knowing which law applies and having the resources to pursue whoever deployed the harmful model. He places responsibility on regulators to identify the relevant law and determine how to enforce it.
Companies must participate because they understand their technologies and limitations, he argued. Yet competition and the cost of safeguards make them unlikely to lead the creation of laws and regulations. That leadership, in his view, must come from government—not from leaving developers to set the terms themselves.
The public register remains a work in progress and is updated monthly. Its continuing task is to track both sides of the problem: newly identified harms and the defenses available—or still needing development—to protect the people affected.
Loading discussion...
Join the conversation
Explain which approach would better protect people harmed by AI.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.