World Solve

Consumer-grade detection tools for hidden surveillance devices remain unreliable and often expensive, leaving many people vulnerable to covert monitoring in their own homes or workplaces. Making effective detection accessible to everyday consumers remains unresolved.

Insurance companies increasingly use AI to assess claims, but there's limited independent, scalable auditing to ensure these systems aren't unfairly denying or delaying valid claims. Building trustworthy oversight for this remains an unresolved regulatory and technical challenge.

Pirated or counterfeit software remains a common vector for malware distribution, and current detection and prevention methods haven't fully closed this persistent security gap. This continues to put both individual users and organizations at unnecessary risk.

Voice and text-based interfaces for controlling software are improving, but ensuring they interpret complex commands accurately and safely, without unintended consequences, remains unresolved. This limits how much control can be safely handed over to natural language alone.

AI can now generate highly personalized, convincing spam and phishing messages faster and at greater scale than ever before, outpacing many existing filtering systems. Keeping filters effective against this rapidly evolving threat remains an ongoing challenge.

Insurers offering cybersecurity coverage often struggle to accurately assess a company's true risk level, leading to mispriced policies that either overcharge safe companies or undercharge vulnerable ones. Better, more accurate risk assessment tools remain underdeveloped.

Many critical systems, including in banking and government, still run on decades-old code that few current developers understand, and automated modernization tools remain unreliable for anything but simple cases. This leaves critical infrastructure dependent on shrinking expertise.

Even after an AI model is retired, techniques exist that can sometimes extract fragments of its original training data, posing ongoing privacy risks. Fully preventing this kind of data leakage after decommissioning remains unresolved.

As AI-powered search and assistants become primary ways people find information, a small number of companies effectively shape what most people see and believe. Ensuring this influence remains fair and transparent remains a largely unresolved governance issue.

Low-code and no-code platforms have made software creation more accessible, but users without security expertise can still inadvertently build applications with serious vulnerabilities. Making these tools both accessible and inherently safe remains an unresolved design challenge.

AI hiring systems trained on historical hiring data can inadvertently learn and repeat past patterns of discrimination, even when explicitly designed to avoid it. Fully eliminating this bias while maintaining useful screening capability remains unresolved.

A surprising amount of the internet's daily traffic runs through a small number of major cloud providers, meaning a single technical failure can take down countless unrelated websites and services at once. Reducing this concentrated fragility remains an unresolved structural issue.

Video files, especially in high resolution, still require significant storage and bandwidth despite decades of compression research, and further meaningful improvements have become harder to achieve. This constrains streaming quality and cost, especially in areas with limited bandwidth.

As AI-generated faces and voices become more realistic, some biometric security systems can potentially be fooled by convincing synthetic versions of a real person's face or voice. Strengthening these systems against this emerging threat remains an active challenge.

Automatically applying every software update reduces security risk but occasionally introduces new bugs or incompatibilities, while manual review introduces dangerous delays. Balancing these risks reliably remains an unresolved operational challenge.

Techniques that let multiple parties jointly compute something without revealing their private data to each other exist, but they remain too slow and complex for widespread practical adoption. Making this both secure and fast enough remains an active research challenge.

While hardware energy use gets some attention, the carbon footprint of running specific pieces of software, from development to daily use, remains largely unmeasured and unaccounted for. Building this tracking into standard practice remains unresolved.

People increasingly turn to AI chatbots for guidance on legal or financial questions, but these systems can provide plausible-sounding yet incorrect advice with no clear warning of their limitations. Better safeguards around this kind of high-stakes advice remain underdeveloped.

Harvesting delicate crops still often requires human dexterity and judgment that current agricultural robots can't consistently replicate at an affordable cost. Solving this could ease labor shortages in farming, but remains technically unresolved.

Companies increasingly build products using AI models developed by others, but there's no consistent, scalable way to verify those models are safe and reliable before integration. This creates hidden risk throughout the growing AI supply chain.