World Solve

Climate change and urban pumping changes may cause groundwater to enter basements, transit systems, utilities, and foundations. Cities need mapping, waterproofing, drainage, and financing strategies that address slow cumulative damage. The design must remain functional when standards, trade routes, data access, and diplomatic cooperation are unreliable. The research opportunity is to define measurable success criteria, test the approach in realistic settings, identify failure modes, and create a pathway from prototype to accountable deployment.

Automated trading algorithms can sometimes trigger rapid, extreme price swings in smaller or less liquid markets, especially during unusual conditions, without any single actor intending market harm. Building safeguards against this kind of systemic risk remains incomplete.

Safety filters designed to block clearly harmful requests can sometimes be bypassed through indirect or creatively phrased questions that still elicit dangerous information. Closing this gap while preserving the system's general usefulness remains an unresolved balancing act.

Attackers can automatically test massive numbers of previously leaked username and password combinations across many websites to find accounts where the same credentials still work. Reducing the effectiveness of this widespread technique remains an unresolved security challenge.

Automated pricing systems can sometimes raise prices sharply during shortages or emergencies based purely on demand signals, without regard for fairness or necessity. Building ethical guardrails directly into these systems remains inconsistently implemented.

Coordinated groups, sometimes aided by bots, can flood a business or creator with negative reviews for reasons unrelated to actual quality, and platforms struggle to distinguish this from genuine customer dissatisfaction at scale. Fair, reliable detection of this manipulation remains unresolved.

A fabricated but realistic-seeming weather alert or disaster warning could spread panic or cause people to take unnecessary risky action before it's identified as false. Faster detection and correction systems for this kind of misinformation remain underdeveloped.

Employees or attackers can sometimes use carefully crafted prompts to trick AI systems integrated into company tools into revealing sensitive internal information they weren't meant to share. Fully closing this security gap remains an unresolved AI safety challenge.

AI-generated medical images used for testing or training purposes could potentially be mistaken for real patient data if not carefully labeled and verified, creating risk of diagnostic confusion. Building reliable safeguards against this remains an emerging challenge.

Even without directly stating private information, AI models can sometimes reveal sensitive details indirectly through patterns in how they respond to certain prompts. Fully closing this subtle privacy leak remains an unresolved technical challenge.

AI can now craft phishing emails with perfect grammar and highly personalized context that closely mimic real business correspondence, making them far harder for both people and filters to catch. Staying ahead of this improving deception remains an unresolved challenge.

A small number of companies control the primary channels through which most people discover and install mobile apps, giving them significant power over which businesses can reach customers at all. Balancing platform control with fair competition remains an unresolved policy issue.

Data brokers can aggregate information from many different sources to build surprisingly detailed profiles on individuals, often without those individuals ever being aware or having consented. Meaningful regulation and technical limits on this practice remain inconsistent globally.

AI can generate large volumes of convincing, varied text quickly, making it easier for bad actors to run coordinated disinformation campaigns that are harder to detect than repetitive, obviously automated posts. Staying ahead of this evolving tactic remains an unresolved challenge.

Advanced surveillance technology developed for legitimate security purposes can end up being sold to and misused by governments to suppress dissent. Preventing this kind of misuse through export controls or corporate accountability remains inconsistently enforced.

AI can now generate entirely fake but convincing photos and personas used to build fraudulent romantic or social relationships online for financial exploitation. Detecting and removing these fake profiles quickly enough remains an unresolved platform challenge.

Scammers can deploy AI-driven chat systems that convincingly impersonate real customer service representatives to extract sensitive information from victims. Detecting and shutting down these operations quickly remains an unresolved challenge for platforms and regulators.

Cookie consent banners have become a near-universal nuisance without necessarily giving users meaningful, easy-to-understand control over their data. Redesigning this system to be both genuinely protective and user-friendly remains unresolved.

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.

Some of the internet's most widely used open-source components are maintained by very few volunteers, and serious vulnerabilities have been found lurking undetected for years in code millions of systems depend on. Better resourcing and review processes for open-source security remain inconsistently funded.