Hi

Welcome (back) to The Prompt EU. AI isn't well understood, but we learn a lot in our work that can help. In this newsletter, we share some of those learnings with you. If you find them helpful, make sure you're signed up for the next issue.

[News] EU Cyber Action Plan

As frontier AI becomes more capable in cybersecurity, access to defensive tools matters. OpenAI’s newly launched EU Cyber Action Plan is focused on making advanced cyber models available to trusted European defenders, including EU Institutions, Member State cyber authorities, and companies in critical sectors.

At OpenAI we’ve been preparing for years to advance the cyber capabilities of our AI models, including through the establishment of our Preparedness Framework in 2023 which covers cybersecurity risks. Over the past months, we’ve doubled down on our efforts to help make advanced cyber AI capabilities available for many defenders, not just the few.

Through our Trusted Access for Cyber program, more than a dozen European companies such as Deutsche Telekom and BBVA have already accessed OpenAI’s latest cyber models, including GPT-5.5-Cyber, for defensive workflows, such as vulnerability identification, remediation, and incident response. The goal is to help protect Europe’s institutions, infrastructure, companies, and citizens, while ensuring powerful cyber capabilities do not remain concentrated in the hands of a few actors. Read our five-part action plan here.

[News] Malta’s AI for All

OpenAI and the Government of Malta announced a world first partnership to provide ChatGPT Plus access to all citizens. Subject to completing the locally designed AI for All training course by the University of Malta, the initiative offers practical AI literacy at a national level and a route to access ChatGPT Plus for one year at no cost to eligible participants.

“Malta is leading the way by showing how countries can empower their citizens to benefit from the transformative potential of AI,” said George Osborne, Head of OpenAI for Countries. “Intelligence is becoming a national utility and governments have an important role to play in making sure people have both access to the tools and the skills to use them well. Malta is putting that into practice nationwide so AI can support people in their everyday lives, at work and across their communities.”

For Europe, Malta offers a concrete model for national AI adoption, combining access and skills rather than access alone. As governments look for practical ways to help people use AI responsibly at work, in learning, and in everyday life, the Malta approach shows how governments can work with partners like OpenAI to support responsible adoption, pairing AI literacy with access to advanced AI tools.

[Blueprint] European Youth Safety Blueprint

OpenAI published its European Youth Safety Blueprint, setting out five pillars for policymakers working to protect young people while preserving access to AI tools that support learning, creativity, and opportunity. The blueprint focuses on responsible AI adoption in education, age-appropriate experiences, under-18 safety policies, protections against manipulative or deceptive AI outputs, and accessible parental controls.

This follows our global Child Safety Blueprint published last month, setting out a framework to combat AI-enabled child sexual exploitation.

Taken together, the two blueprints set out a practical approach to youth safety in Europe: pairing clear rules and strong platform safeguards with continued access to AI tools that can support learning, creativity, and opportunity.

[Tools] OpenAI Privacy Filter

OpenAI released Privacy Filter, an open-weight model designed to detect and redact personally identifiable information in text. It can run locally, process long inputs efficiently, and identify context-dependent personal data that traditional pattern-matching tools often miss.

We built Privacy Filter to make privacy protection easier for developers to integrate directly into AI development workflows. For teams working with sensitive data, including in Europe, that means stronger safeguards can be part of how AI systems are developed and evaluated from the start.

[Paper] Ideas to Keep People First

We published Industrial Policy for the Intelligence Age: Ideas to Keep People First to start a broader conversation about how governments, companies, civil society, and communities can prepare for the economic and social changes that more capable AI systems may bring. Our view is that AI policy should focus not only on innovation, but also on broad access, shared prosperity, worker voice, modern safety nets, and resilience against frontier risks.

For Europe, the conversation connects closely to ongoing debates about competitiveness, skills, and how to make AI adoption work for people in practice. As the EU moves from setting rules to enabling uptake across the economy, we believe industrial policy can help ensure that AI-enabled growth reaches workers, small businesses, communities, and public services, not only the largest firms.

[Tools] Codex is for Everyone

A recent OpenAI Forum session looks at how Codex can help people who are not software developers explore ideas, build lightweight tools, and turn expertise into something others can use. Give it a try!

[Disclosure]

Today’s newsletter and images co-created with ChatGPT.

Hi

Welcome (back) to The Prompt EU. AI isn’t well understood, but we learn a lot in our work that can help. In this newsletter, we share some of those learnings with you. If you find them helpful, make sure you’re signed up for the next issue.

This edition looks at how Europe’s AI debate is moving from strategy to real-world deployment. From kicking off our European SME AI Accelerator in Dublin, to national efforts like Greece’s system-wide approach, the focus is shifting toward practical adoption and competitiveness. At the same time, new tools and frameworks — from parental guidance to learning outcome measurement and interactive education features — reflect a broader effort to make AI both more useful and more understandable in everyday contexts. What we’re seeing: AI in Europe is entering a more operational phase, where impact is defined not by access, but by how effectively it is used.

[News] Launching our SME AI Accelerator

The first European SME AI Accelerator kicked off in Dublin last week, gathering over 130 small businesses at Dogpatch Labs, alongside our other partners Booking.com, Retail Excellence Ireland and The Innovation Exchange. Throughout the day, SMEs worked hand-in-hand with OpenAI experts to go beyond the basics — and see how AI can really help transform their businesses.

Our new research across 4,700 SME decision-makers shows AI is already mainstream, with eight in ten using it at work, and delivering real productivity gains of over five hours saved each week. But adoption remains uneven, with fewer than half using it regularly and many businesses still lacking the skills and confidence to go further.

That gap is exactly what the programme is designed to address. Over the next year, we will support 20,000 SMEs through hands-on workshops, online bootcamps and practical, business-focused use cases.

The programme will expand to France, Germany, Italy, Poland, Ireland and the UK, with a clear aim: helping European small businesses turn AI into a genuine competitive advantage.

[Insight] AI and influence operations: evolution, not revolution

In a recent interview, our investigator Ben Nimmo shares how we are seeing AI shape influence operations in practice. The picture is incremental rather than disruptive: AI is being integrated into existing workflows, making familiar tactics faster — particularly in how content is generated and iterated — while the underlying infrastructure of influence operations remains largely unchanged.

At the same time, capabilities are shifting on the defensive side. We are using AI to analyse large volumes of data, identify patterns, and investigate campaigns far more quickly, and these tools are increasingly accessible to users themselves, who can rely on them to assess suspicious content in real time. Overall, while tactics remain familiar, the ability to analyse and respond to them is improving significantly.

[Study] Can AI help Europe leapfrog? Greece is testing it

A new Harvard Business School case looks at Greece’s attempt to use AI not as a standalone tool, but as a system-level lever for education, entrepreneurship, and growth. Rather than piloting at the margins, the model combines teacher-led AI integration in schools, a national startup accelerator, and public-private delivery to move quickly despite institutional constraints.

The bet is clear: by building AI literacy early and linking it directly to venture creation, countries like Greece could bypass structural bottlenecks — from weak education outcomes to talent outflows. What we’re seeing: a more operational model of “AI for competitiveness” emerging — less about standalone regulation, and more about how to actually deploy AI across institutions at speed.

[Resource] Parental Hub

We’ve launched a new parent resource hub to support how families engage with AI at home. As more teens use tools like ChatGPT for school and everyday questions, the goal is to give parents clear, practical guidance — how the technology works, where its limitations are, and how to have informed conversations about its use. The hub builds on existing parental controls, with a strong focus on helping families actively shape safe and constructive use.

Alongside product safeguards and policy discussions, there is a growing need for accessible, real-world guidance that helps people navigate these tools in practice. By combining protections with literacy, our aim is to help ensure families have both the tools and the understanding to use AI responsibly.

[Insight] Measuring AI in Education

We’ve introduced a new Learning Outcomes Measurement Suite, developed with partners including the University of Tartu and Stanford’s SCALE Initiative, to better understand how AI shapes learning over time. The goal is to move beyond narrow signals like test scores and capture how students actually engage with AI in real-world settings.

The framework enables longitudinal measurement across model behaviour, learner interactions and cognitive outcomes — including engagement, persistence and critical thinking — and is already being studied at scale with students in Estonia. What we’re seeing: as AI adoption accelerates, the focus is shifting from short-term gains to building a more robust evidence base on how AI shapes learning itself.

[Prompt] Exploring concepts in real time

We’re introducing interactive math and science tools in ChatGPT to support more active learning, allowing users to explore and experiment with concepts in real time — just try a prompt like “walk me through the ideal gas law” to get started.

[Podcast] Inside our approach to the Model Spec

We shared new insights into our Model Spec — the framework guiding how models follow instructions, resolve conflicts, respect user freedom, and behave safely across a wide range of queries. The full conversation is below.

[Disclosure]

Today’s newsletter and images co-created with ChatGPT.

Hi

Welcome (back) to The Prompt EU. AI isn’t well understood, but we learn a lot in our work that can help. In this newsletter, we share some of those learnings with you. If you find them helpful, make sure you’re signed up for the next issue.

Europe enters the next phase of the AI transition with a paradox: strong foundations in talent, research, and values, yet a persistent gap between AI potential and how well AI tools are utilised. While interest in AI is high, usage remains uneven across countries, sectors, and firm sizes, and many organisations are still struggling to translate technological capability into productivity and growth. Closing this gap is becoming a central economic and policy challenge for Europe as it looks to compete, innovate, and deliver shared benefits from AI.

Against this backdrop, OpenAI has announced a set of new initiatives to support Europe’s next chapter in AI adoption. These include the EU Economic Blueprint 2.0, a new SME AI Accelerator to help smaller businesses build practical AI capabilities, expanded country-level partnerships through OpenAI for Europe, and a new €500,000 NGO grant to support research into youth safety and wellbeing. In this edition of The Prompt EU, we take a closer look at each of these announcements — what they aim to address, how they fit together, and what they could mean for Europe’s AI future.

[Report] EU Capability Overhang

AI capabilities are advancing fast, but their economic impact depends on how deeply they’re used in practice. Globally, there is a large “capability overhang”: while frontier AI can now handle complex, multi-step tasks, most users still rely on simpler interactions. The data shows the scale of this gap — globally, power users (95th percentile) use roughly 7× more thinking capabilities than the typical user — pointing to significant unrealised productivity potential.

Europe performs comparatively well, but unevenly. On average, EU countries use 17% more thinking capabilities per person than the global baseline, yet 9 Member States still fall below the global average, and the spread between the most and least AI thinking-intensive EU countries is around 40%. The chart highlights distinct national patterns: some countries pair high penetration with deep use (notably Lithuania and Latvia), others show deep use among a narrower base (such as Poland, Greece, and Italy), while a third group has broad uptake but lower average depth (including the Netherlands and Luxembourg).

What matters next is not just more adoption, but more effective adaption. Europe should help mainstream users move from surface interactions to higher-value workflows; scale what already works in leading pockets while closing gaps elsewhere; and invest in practical skills, sector-specific use cases, and trusted deployment so deeper AI use translates into productivity and growth. Read the full report here.

[News] SME AI Accelerator for Europe

Closing Europe’s AI capability gap also requires action beyond policy. According to the latest Eurostat figures for 2025, AI adoption among small businesses stands at just 17%, compared with 55% among large enterprises. If over the years, this gap keeps widening, it will have real implications for productivity and competitiveness of the continent. To help close it, OpenAI is launching a new SME AI Accelerator, in partnership with Booking.com, to support 20,000 SMEs across the economy in boosting productivity and growing their businesses with AI. The programme will roll out across France, Germany, Italy, Poland, Ireland, and the UK, and is open to small business owners and teams in all sectors, including those with no technical background. It will combine in-person workshops with virtual training delivered through OpenAI Academy, with more details to follow soon.

[Grant] EMEA Youth & Wellbeing Grant

Alongside adoption, trust remains a precondition for AI’s success in Europe, especially for the next generation growing up with AI as part of everyday life. To support evidence-based approaches to youth protection and wellbeing, OpenAI has announced a €500,000 EMEA Youth and Wellbeing Grant, aimed at strengthening collaboration between youth organisations and research institutions. The programme will fund projects focused on youth protection, digital wellbeing, and practical research that can inform safer AI use for young people. We strongly encourage eligible organisations and researchers to apply — details on the scope, eligibility, and application process are available here.

[News] Our Approach to Age Prediction

OpenAI is rolling out age prediction on ChatGPT consumer plans to help estimate whether an account likely belongs to someone under 18, so the right experience and safeguards can be applied — while still allowing adults to use the tools as intended. The system looks at different behavioral and account-level signals linked to your account, for example, it may look at general topics you talk about or the times of day you use ChatGPT. When it flags an account as likely under 18, it automatically applies additional protections to reduce exposure to sensitive or potentially harmful content. If someone is incorrectly placed in the under-18 experience, they can quickly confirm their age (via a selfie check through Persona) and restore full access. In the EU, age prediction will roll out soon to account for regional requirements. To learn more please see here.

[News] Education for Countries

​​If Europe wants AI to translate into long-term productivity and opportunity, it needs to start where skills are formed: in classrooms, universities, and lifelong learning systems. Through Education for Countries, OpenAI is partnering with governments and university consortia to support the responsible integration of AI into education — helping personalise learning, reduce administrative burden for educators, and prepare students for the workforce. The initiative is already underway with European partners including Estonia, Greece, Italy (CRUI), and Slovakia, combining AI tools for learning, certifications and training, and collaboration on national research into learning outcomes and teacher productivity. If you’d like to explore the approach — and what these early European partnerships are testing in practice — you can find more details here.

[OpenAI Forum] Using AI to Protect Children Online

To continue the discussion on youth safety and responsible AI deployment, we’re hosting a Forum conversation with Thorn on how AI can be used to help protect children online. The session will explore practical applications of AI in combating online harms, the role of collaboration between technology developers and child-protection organisations, and what effective safeguards look like in practice. We warmly invite you to join the conversation on Wednesday, 11 February at 6:00pm CET— you can find full details and register here.

06:00 PM – 06:50 PM CET on Feb 11

Register

[Quick Links]

• OpenAI Privacy Hackathon Report: Following the success of our first OpenAI Privacy Hackathon, we’ve published a report capturing the key takeaways from the event focused on turning privacy ideas into real-world impact. More info here.

• Keeping your data safe when an AI agent clicks a link: AI systems are getting better at taking actions on your behalf, opening a web page, following a link, or loading an image to help answer a question. These useful capabilities also introduce subtle risks that we work tirelessly to mitigate. Read more here on how we’ve built safeguards to reduce the risk when ChatGPT (and agentic experiences) retrieve web content.

[Disclosure]

Today’s newsletter and images co-created with ChatGPT.

Hi

Welcome (back) to The Prompt EU. AI isn’t well understood, but we learn a lot in our work that can help. In this newsletter, we share some of those learnings with you. If you find them helpful, make sure you’re signed up for the next issue.

Much has been said this year about the work Europe needs to do on AI adoption and that despite strong talent, European enterprises have been slower to translate AI investment into productivity and growth. That picture is starting to shift.

Over the past months, we’ve seen major European companies move decisively from experimentation to deployment. Our recent collaborations with BBVA and Deutsche Telekom show what large-scale, responsible AI adoption looks like in practice—inside highly regulated, pan-European organisations. At the same time, our new ‘The State of Enterprise AI Report’ points to clear momentum across Europe, with countries like France and the Netherlands among the fastest-growing enterprise markets worldwide and Germany among the largest outside the U.S.

This matters because 2026 will be a key year. More European businesses are knuckling down on real adoption—integrating AI into everyday workflows, not just pilots—while EU policymakers begin the vital work of simplification: making sure the regulatory framework supports competitiveness, investment, and deployment at scale. Getting this right is essential if AI is to translate into productivity gains for Europe’s economy.

In this edition of The Prompt EU, we share what we’re seeing on the ground—from enterprise adoption and scientific research, to safety practices and real-world deployments—so you can track how AI is moving from promise to practice across Europe.

[Report] The State of Enterprise AI

OpenAI’s economics research team has released ‘The State of Enterprise AI Report,’ a comprehensive look at how enterprises are adopting AI. The report shares what workers say they’re gaining, and how organizational leaders are turning experimentation into measurable productivity and new capabilities, and details Europe’s role among global adoption and usage at work.

Key findings:

1. AI is moving from chatbots and experimentation to core infrastructure in work.
Usage is up and deepening. An 8x rise in enterprise message volume and 19x growth in structured workflows (Custom GPTs, Projects) show companies starting to embed AI into repeatable processes. Firms are beginning to reorganise work around AI, not just ask it occasional questions.

2. Enterprises that leverage AI are experiencing measurable productivity and business impact.
Enterprise users report saving 40–60 minutes per day and being able to complete new technical tasks such as data analysis and coding. 75% of respondents report faster or higher-quality work, with many saying that AI now lets them complete tasks they couldn’t previously do, including coding and analysis. Yet, a sizeable share of employees haven’t touched core features like data analysis or reasoning.

3. A widening gap is forming between AI “frontier” firms/workers and everyone else.
Frontier employees send 6x more messages and engage across far more task types; frontier firms integrate AI into workflows 7x more deeply.

4. Europe is central to the global picture. European enterprises aren’t lagging — they’re helping shape what real adoption looks like.
France and the Netherlands are among the fastest-growing enterprise markets worldwide; the UK and Germany now rank among the largest outside the U.S.

This report and analysis draws on two novel sources of data (deidentified and aggregated to preserve privacy):

• Real-world usage data from among the 1 million business customers of OpenAI.

• An OpenAI survey of 9,000 workers across almost 100 enterprises documenting patterns of AI adoption.

Please find the report here and blog here.

[News] OpenAI Expands Strategic Collaborations with BBVA and Deutsche Telekom

Our recent collaborations with BBVA and Deutsche Telekom offer a concrete view of how AI is already being deployed at scale in highly regulated sectors — and what that looks like in practice.

OpenAI is working with BBVA to advance its AI strategy and advance its ambition to become an AI-native bank. BBVA is giving access to ChatGPT Enterprise to its global workforce of 120,000 employees. The focus is on supporting productivity, internal knowledge sharing, and employee upskilling and improving customer experiences, while operating within established security, data protection, and governance frameworks. This builds on a successful program that saw 11,000 users saving almost three hours a week on routine tasks, with over 80% engaging daily.

In parallel, Deutsche Telekom’s collaboration with OpenAI will bring AI to millions of people across Europe with multilingual access, privacy-first deployment, and trust. Employees will get access to AI tools and training to help teams work more effectively, improve customer service, and spend more time on innovation. OpenAI will also work with the major European infrastructure provider to develop privacy-first consumer-facing applications, and processes to aid customer support and optimize network operations.

Data from the State of Enterprise AI Report and these two collaborations demonstrate that European enterprises aren’t lagging — they’re helping shape what real AI adoption and business innovation looks like.

[Insight] Strengthening Safety with External Testing

In our latest blog post, we lay out how we use independent, trusted third-party assessments to stress-test frontier models—validating key safety claims, reducing blind spots, and increasing transparency around capabilities and risk.

In practice, our external testing work comes in three forms: independent evaluations by external labs (across areas like biosecurity, cybersecurity, AI self-improvement, and scheming), methodology reviews (external scrutiny of how we evaluate and interpret risk), and subject-matter expert probing (experts testing real workflows and feeding structured input into deployment decisions).

We’ll keep investing in external testing because it brings additional rigor and perspective to our safety work—and helps ensure deployment decisions are grounded in evidence, not assumptions.

[Model Spec] Reinforcing Teen Protections

OpenAI has updated its Model Spec to formalize a dedicated set of Under-18 Principles, grounded in developmental science and informed by external expert input. The update explains how our core safety rules apply in adolescent contexts and is organized around four commitments: put teen safety first; promote real-world support; treat teens like teens; and be transparent. These safeguards build on the Model Spec that governs behavior for all users and complement product-level protections for teenage users, including parental controls. Read the full update here.

[Podcast] Accelerating Science

Frontier AI is starting to show what it can mean for scientific research in practice. If you’d like a quick, first-hand feel for how scientists are using frontier models in day-to-day work, this short podcast brings their experiences to life.

[Disclosure]

Today’s newsletter and images co-created with ChatGPT.

Hi

Welcome (back) to The Prompt EU. AI isn’t well understood, but we learn a lot in our work that can help.

This week marks a pivotal moment for Europe’s tech future as the EU Member States set the course on digital sovereignty and the European Commission launches the Omnibus to simplify the digital rulebook. These debates mirror the discussions we had last week across Europe and the concrete steps we’re taking to move from recommendations to action, all outlined in this edition. If you find these insights helpful, make sure you’re signed up for the next issue.

[News] Announcing OpenAI for Ireland

Sovereignty debates are usually focused on the tech stack but that will mean nothing without having the talent to actually build on and use the technology. Last week, we launched OpenAI for Ireland, a new initiative to help Irish SMEs, startups and young builders harness AI for growth working with the Irish Government, Dogpatch Labs and Patch. The programme includes an SME Booster launching in 2026 with the Department for Enterprise, Tourism and Employment⁠, offering national AI skills training, workshops and mentoring; a partnership with Dogpatch Labs to help early-stage founders integrate AI into products; and a three-year collaboration with Patch to expand its summer programme, fellowships and grants for young builders. This comes as Ireland deepens its role as a tech hub, with over one million people using ChatGPT weekly.

As the Government refreshes its National AI Strategy and prepares to host the EU AI Summit during its Council Presidency in 2026, OpenAI will continue supporting Ireland’s ambition to embrace AI as the next major technological leap. Through sustained investment, partnerships and skills programmes, the goal is clear: broaden AI adoption and ensure the benefits of this technology are felt across the country.

[News] OpenAI Signs Declaration on Youth Protection in the Age of AI

Last week in Rome, Jason Kwon, our Chief Strategy Officer, joined the High-Level Meeting on Dignity of Children and Adolescents in the Age of Artificial Intelligence hosted by the Fondazione Child and the Child Dignity Alliance and attended by faith leaders, youth experts and industry representatives to discuss responsible AI innovation. OpenAI signed a multi-stakeholder declaration, presented to Pope Leo XIV, calling for a global commitment to protect children in the age of AI while enabling them to benefit from the opportunities AI unlocks. As Jason noted, today’s teens will be the first generation to grow up with AI woven into daily life; getting this right requires thoughtful guardrails that build trust and empower young people to learn, create and thrive.

[Blueprint] Protecting Teen ChatGPT Users

OpenAI released its first Teen Safety Blueprint—a policy framework outlining how AI systems should be designed for young people, grounded in safety-by-design principles, controls and age-appropriate experience. The blueprint details our measures such as privacy-preserving age prediction that defaults to an under-18 experience when we are not confident about someone’s age or have incomplete information; parental controls to give families choices; and research partnerships to better understand healthy AI use.

[News] Our Partnership with Techleap

OpenAI launched a new partnership with Techleap, the Netherlands’ leading startup platform, to support the next generation of Dutch AI founders. As a new accelerator node within OpenAI for Countries, the collaboration will give a selected cohort of startups access to API credits, technical support and our latest frontier models. The partnership will help Dutch AI startups experiment, prototype, and scale, while fostering a strong, founder-led hub for safe and responsible AI development. It also reinforces the Netherlands’ position as one of Europe’s most dynamic, innovation-driven AI ecosystems in Europe. This marks the start of a long-term collaboration with the wider Techleap community to support ambitious founders and help entrepreneurs build globally competitive AI products from the Netherlands.

[Event - AI Privacy Hackathon]

On 4 December 2025 in Dublin, we’re hosting our first AI Privacy Hackathon, bringing together developers, engineers, and privacy experts to co-create practical, privacy-enhancing AI solutions and help build a strong community of practice.

If you’d like to join, please fill out this form with a short note on why you’d be a good fit, and we’ll review your request.

[Event Recording]

Jason Kwon recently discussed OpenAI’s updated structure and the launch of the OpenAI Foundation, explaining how this strengthens our ability to advance frontier AI safely while accelerating work on health breakthroughs and practical safety solutions. Catch the full discussion here.

[Quick Links]

• Aardvark: We’ve introduced Aardvark, an agentic security researcher powered by GPT-5 that autonomously scans, reasons about and helps fix software vulnerabilities—now in private beta. Learn more here.

• Keeping Users Safe in the Age of AI: Supporting young people and their families with safety systems, parental controls and learning environments is a core priority for OpenAI. Learn more here.

[Disclosure]

Today’s newsletter and images co-created with ChatGPT.

Hi

Welcome (back) to The Prompt EU. AI isn’t well understood, but we learn a lot in our work that can help. In this newsletter, we share some of those learnings with you. If you find them helpful, make sure you’re signed up for the next issue.

[Response] Harmonizing Europe’s digital rulebook for the AI opportunity

Europe’s ability to lead in the AI era depends not only on investment and innovation, but also on clear, aligned, and proportionate rules. Although designed to protect citizens and foster trust, the cumulative effect of all the digital laws has introduced complexity, raising costs for companies and slowing diffusion of innovation across the Single Market. The upcoming Digital Omnibus Package offers a rare opportunity to make the EU’s digital rulebook more agile and innovation-friendly.

Our contribution to the EU’s Digital Simplification call for evidence sets out five pragmatic pillars for action: horizontal simplification to streamline overlapping digital laws; vertical simplification to bridge gaps between EU and national implementation; international convergence to align regulatory frameworks with democratic partners; single market acceleration to integrate data, compute, capital, and talent; and AI itself as a tool for smarter governance. Simplification of Europe’s digital rulebook is not about deregulation, but about smart regulation, agile implementation and technological facilitation.

[News] OpenAI for Germany: frontier AI, on German terms

Last month, OpenAI and SAP announced the launch of OpenAI for Germany, a partnership to bring SAP’s enterprise applications expertise and OpenAI’s leading AI technology to Germany’s public sector.

The collaboration will enable millions of public sector employees to use AI safely and responsibly while meeting strict data sovereignty, security, and legal standards. The ambition is to bring AI into the heart of administrative workflows, automating tasks, improving process efficiency, and allowing civil servants to focus more on public service and less on paperwork, supporting the Federal Government’s State Modernization ambitions and Hightech Agenda’s target of achieving AI-driven value creation equivalent to 10% of GDP by 2030.

To ensure sovereignty, OpenAI for Germany will be supported by SAP’s subsidiary Delos Cloud, running on Microsoft Azure technology. SAP plans to expand the existing infrastructure of Delos Cloud in Germany to 4,000 GPUs for AI workloads to support this effort.

[Poll] How Europeans Use AI

Our recent poll across seven EU countries (France, Germany, Belgium, Poland, Ireland, Denmark, Italy) shows that 80% of the informed public use AI tools daily or weekly, underscoring how quickly AI has become part of everyday routines. In Brussels, usage grew particularly fast — up 14 percentage points in the past year to reach 93% of respondents using AI at least weekly.

The top reasons people use AI are to access information quickly (54%), learn new things (41%), translate and summarize information (39%), increase productivity (38%), and generate creative ideas (38%). These figures show that AI is already a trusted, everyday tool for learning, translation, problem-solving, and creativity across Europe.

[News] Parental Controls + Wellbeing Council

We’ve rolled out parental controls and a new parent resource page to help families guide how ChatGPT works in their homes. Available to all ChatGPT users, parental controls allow parents to link their account with their teen’s account and customize settings for a safe, age-appropriate experience. Parents can customize their teen’s experience in ChatGPT in a simple control page in account settings. If a teen unlinks their account, their parent will be notified. Find out more in our Parental Resource Hub.

We’ve also assembled the Expert Council on Well-Being and AI to help guide our ongoing work. The eight-person council brings together leading researchers and experts with decades of experience studying how technology affects our emotions, motivation, and mental health. Their role is to advise us, pose questions and help define what healthy interactions with AI should look like for all ages.

[Event] AI & Youth — Academic Roundtable

On October 28, 2025 in Brussels, in partnership with ICLE, we’re hosting an in-person, closed-door roundtable with our product experts to dive into the intersection of AI and youth.

Are you conducting research on this topic and want to join the conversation? Fill out this form to learn more.

[Event] Hacktivate AI — Thank You For Joining!

Thank you to everyone who joined us for Hacktivate AI in Brussels, where 65 leaders from across EU institutions, governments, startups, academia, and civil society came together in a full-day policy hackathon. The output is a report of 20 actionable ideas to accelerate AI adoption in Europe, spanning workforce upskilling (with the Individual AI Learning Account), SME support via an AI Champions Network, shared public-sector infrastructure through a European GovAI Hub, and a strong emphasis on “relentless harmonisation” to ensure regulatory alignment.

[Quick Links]

• Defining and Evaluating Political Bias in LLMs: We’ve introduced a new framework to assess how language models respond to politically sensitive prompts, showing that bias is both rare and decreasing in our latest models. Learn more about the research here.

• Privacy Filter: Find out more about Privacy Filter — our privacy tool built with our most advanced language models to detect and mask personal information in training data. Read more here.

[Disclosure]

Today’s newsletter and images co-created with ChatGPT.

Hi

Welcome (back) to The Prompt EU. AI isn’t well understood, but we learn a lot in our work that can help. In this newsletter, we share some of those learnings with you. If you find them helpful, make sure you’re signed up for the next issue.

[News] OpenAI for Greece

Greece has become one of the first countries to partner with OpenAI on a national program, “OpenAI for Greece.” Together with the Greek Government, the Onassis Foundation, and Endeavor Greece, we’re rolling out ChatGPT Edu in secondary schools to strengthen AI literacy and help teachers and students use these tools responsibly.

The initiative also supports Greece’s innovation ecosystem—offering startups access to OpenAI tools, mentorship, and global exposure—while aligning with national priorities in education, public services, healthcare, and climate. From Plato’s Academy to Aristotle’s Lyceum, Greece is the historical birthplace of western education. It is now opening a new chapter, leading on a new comprehensive effort to prepare citizens, institutions, and entrepreneurs for the opportunities and challenges of the AI era.

[Study] How People Use ChatGPT

OpenAI’s Economic Research team, led by Chief Economist Ronnie Chatterji, and Harvard economist David Deming have published the largest study of ChatGPT consumer usage to date.

Key findings:

• Demographic gaps are shrinking: by July 2025, women comprised 52% of users (up from 37% in early 2024).

• ChatGPT has also become a broadly accessible global tool, with especially rapid growth in low- and middle-income countries. By May 2025, ChatGPT adoption growth rates in the lowest income countries were over 4x those in the highest income countries.

• About 30% of usage is work-related (with writing the top professional task), while 70% is non-work usage—everyday advice, learning, creativity, etc. The study exclusively analyzed consumer users (not enterprise), but highlights that ChatGPT is delivering productivity gains in daily life and for work.

• Patterns of use break down into Asking (49%), Doing (40%), and Expressing (11%)—showing ChatGPT is valued not just for completing tasks, but increasingly as an advisor and decision-support tool.

Together, these results highlight how AI is closing demographic gaps and creating real value for people at work and beyond.

The study was conducted using privacy-preserving methods: researchers did not read user messages, but relied on automated analysis to identify usage patterns. You can read the full paper here.

[Intro] Expanding Economic Opportunity with AI: Matching Skills, Jobs & Access

We’re introducing two new initiatives to help more people benefit from the AI economy. The OpenAI Jobs Platform will connect individuals, small businesses, and local governments with work opportunities in the AI economy, while OpenAI Certifications will make it easier to learn and prove real fluency in using AI tools.

Together with free access through ChatGPT and new learning resources in the OpenAI Academy, our goal is to equip millions of people with the skills they need to thrive in a rapidly changing job market and to give employers confidence in hiring AI-proficient talent.

[Insight] More Helpful ChatGPT

We’re working to make ChatGPT as helpful as possible to people. As part of this, we’re improving how the model recognizes and responds to signs of mental and emotional distress—and, guided by expert input, how it connects people to appropriate care and resources.

We’re also introducing new safeguards for teens and testing age-prediction technology so protections can apply even when sign-up information is incomplete. Parents will be able to link accounts, set controls, disable memory and chat history, and receive alerts if signs of acute distress appear. Stronger default limits and age-appropriate features will help ensure teens have a safer experience from the start.

These updates build on our collaboration with health experts to make ChatGPT more supportive and safe for everyone. We’ll share more about these efforts during our 120-day initiative—learn more here.

[Event] Hacktivate AI - Join Us!

Hacktivate AI comes to Brussels on 23 September (15:00–19:00 CEST). Co-hosted with Allied for Startups, the event will explore how Europe can accelerate AI adoption through policy innovation—bringing together stakeholders from European Institutions, national governments, startups, business, academia, and civil society.

The program features hackathon team pitches, live demos, and a high-level panel on Europe’s path to AI adoption, followed by a networking cocktail. Discover the program and register here.

[Quick Links]

• Privacy Designed with Intelligence: At OpenAI we embed privacy-preserving safeguards into every stage of the AI lifecycle—from pre-training through deployment. Learn more about our approach here.

• Why Language Models Hallucinate: We’ve published new research on why language models hallucinate, exploring the root causes of this behavior and identifying pathways to build more reliable AI systems. You can find the paper here.

[Disclosure]

Graphics created by Base Three using ChatGPT.

Subscribe now