World Solve

A huge share of global internet traffic between continents relies on a relatively small number of undersea cables, and damage to just one can disrupt connectivity for entire regions. Building enough redundancy to fully prevent this remains costly and incomplete.

Fraudsters can use AI to generate entirely fake but convincing identities, complete with fabricated financial histories, to open accounts and commit fraud at scale. Detecting these synthetic identities before damage occurs remains a growing, unresolved challenge for financial institutions.

Ensuring software works well with assistive technologies is often treated as a lower priority than testing for basic functionality or security, leading to inconsistent accessibility even in otherwise well-built products. Making accessibility testing equally rigorous by default remains unresolved.

Age verification systems designed to protect minors from inappropriate content often require handing over sensitive personal information, creating a tradeoff between child safety and privacy for all users. A method that achieves both effectively remains unresolved.

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.

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.

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.

Older adults are often left behind as digital tools and platforms evolve quickly, with educational resources struggling to keep up with both the pace of change and different learning needs. Closing this generational digital literacy gap remains an ongoing challenge.

Even with extensive automated testing, software products regularly ship with bugs that only appear once real users start interacting with them in unpredictable ways. Comprehensive, fully reliable bug detection before release remains an unsolved engineering challenge.

Recommendation algorithms optimize for engagement, which can unintentionally narrow the range of content users are exposed to over time, reinforcing existing beliefs rather than broadening perspective. Designing systems that balance engagement with diversity of exposure remains unresolved.