Seminal
The magazine and atlas of EU-funded researchA new publication from the web and AI team at Scott Charnley

Atlas / United Kingdom

Marie Skłodowska-Curie Action · Running

AURORA Advanced User-centric, Trustworthy and Privacy-preserving AI-powered Search Engine

Coordinator
The Chancellor, Masters and Scholars of the University of Oxford, United Kingdom
Runs
1 November 2026 to 31 October 2028
Status
Running
EU funding
€260k
Total cost
—
Partners
2 organisations in 1 countries
Type
Marie Skłodowska-Curie Action
Grant agreement
101270444

What the project is doing

In the project's own words, from its summary on CORDIS.

The European Green Deal aims to build a resource-efficient, competitive EU economy with net-zero greenhouse gas emissions, clean energy and sustainable industries. Achieving this requires quick and reliable access to information on climate change, energy production, and technological developments. AI-powered search engines (AIPSEs) like Perplexity, can automatically retrieve information from multiple sources and deliver synthesized answers to users—saving them time browsing various websites.

However, AIPSEs are still in their infancy and face critical security challenges, including generating biased or malicious responses, privacy leaks, and limited research in this area. For instance, Google’s AI Overviews generated dangerous responses like suggesting user eat rocks and glue (2024), and attackers can trick AIPSE into citing malicious content from phishing websites. Some AIPSEs track user histories for personalization, raising privacy risks; notably, a ChatGPT glitch exposed users' private conversations (2023).

AURORA (Advanced User-centric, tRustwOrthy and pRivacy-preserving AI-powered search engine) is my proposal to mitigate these challenges, which aims to develop a user-centric AIPSE that provides trustworthy responses while protecting user privacy, and adhering to GDPR and EU AI Act. AURORA includes: 1). a user study on how users perceive malicious responses generated by existing AIPSEs, to guide user-centric design (WP1); 2). a multi-agent debate model with ML-based filters to automatically block fabricated and malicious content (WP2), which is expected to block around 90% of malicious sources, and to reduce hallucinations by about 10%; 3). a local differential privacy scheme to protect user data (WP3).

By preventing the spread of misinformation, blocking malicious responses, and protecting privacy, AURORA enables policymakers, researchers, and citizens to make informed decisions about energy, environmental policies, and sustainable practices.

Topics

Who is involved

OrganisationCityCountryRole
The Chancellor, Masters and Scholars of the University of OxfordOxfordUnited KingdomCoordinator
Imperial College of Science Technology and MedicineLondonUnited KingdomParticipant

Source: CORDIS, European Commission. Data of 5 Nov 2026, reused under the Commission's reuse policy. View on CORDIS · Grant DOI