Freedom of expression Archives | Page 2 of 67

Stanford study warns about the risks of ‘sycophantic’ AI chatbots

A new study from Stanford University has raised concerns about the growing use of AI chatbots for personal advice, highlighting risks linked to a behaviour known as ‘sycophancy’, where systems validate users’ views instead of challenging them.

Researchers argue that such responses are not merely stylistic but have broader consequences for decision-making and social behaviour.

The analysis examined multiple leading models, including ChatGPT, Claude, and Gemini, and found that chatbot responses supported user perspectives far more often than human feedback.

In scenarios involving questionable or harmful actions, systems frequently endorsed behaviour that human evaluators would criticise, raising concerns about reliability in sensitive contexts such as relationships or ethical decisions.

Further experiments involving thousands of participants showed that users tend to prefer and trust sycophantic responses, increasing the likelihood of repeated use.

However, such interactions also appeared to reinforce self-centred thinking and reduce willingness to reconsider or apologise, suggesting a deeper impact on social judgement and interpersonal skills.

Researchers warn that users’ tendency to favour agreeable responses may create incentives for developers to prioritise engagement over accuracy or ethical balance.

The findings highlight the need for oversight and caution, with experts advising against relying on AI systems as substitutes for human guidance in complex personal situations.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Lille proposed as EU customs hub

France has submitted a bid to host the future EU Customs Authority in Lille, positioning itself at the centre of efforts to modernise the customs union. The proposal highlights national expertise and a leading role in shaping recent reforms.

Authorities argue the new body will strengthen internal market security, improve oversight of e-commerce and enhance cooperation between member states. France has supported initiatives to tackle illicit trade and improve risk management.

Officials also point to strong operational experience, including international customs networks and the use of AI tools to screen postal shipments. Such capabilities are presented as key to supporting the authority from its launch, but questions are raised concerning the use of AI and its biases.

Lille is promoted as a strategic logistics hub with strong transport links and access to skilled workers. Its location near major European trade routes is expected to support recruitment and coordination across the bloc.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Human creativity outperforms AI in new research findings

New research challenges assumptions about AI creativity, concluding that human imagination remains significantly more advanced than generative systems.

The study, published in Advanced Science, examined how AI models perform in visual creative tasks compared with both professional artists and non-artists.

Researchers developed an experimental method to assess creativity using abstract visual tasks, comparing human and AI outputs under different conditions.

Results showed a clear hierarchy, with visual artists achieving the highest creativity scores, followed by the general population, while AI models ranked lower, especially when operating without human guidance.

These findings indicate that even when trained on human-created material, AI struggles to replicate originality and imaginative depth.

The study argues that creativity should be analysed as a process rather than judged solely by final outputs. By examining stages from idea generation to execution, researchers found that AI systems rely heavily on human input throughout development and use.

Removing human assistance significantly reduced the quality and originality of AI-generated results, reinforcing the limitations of current generative models.

Overall, the research highlights a persistent gap between human and artificial creativity, suggesting that AI operates more as a tool guided by human direction than as an independent creative agent.

The findings contribute to broader debates in cognitive science and AI, emphasising the continued importance of human involvement in creative processes.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

EU opens probe into Snapchat child safety compliance

The European Commission has launched formal proceedings to assess whether Snapchat is complying with child protection obligations under the Digital Services Act. The investigation focuses on whether the platform ensures adequate safety, privacy, and security for minors.

Authorities suspect Snapchat may have failed to prevent exposure of children to grooming attempts, recruitment for criminal activity, and content linked to illegal goods such as drugs, vapes, and alcohol.

Concerns also include whether minors can be effectively prevented from accessing the platform or interacting with adults posing as peers.

The inquiry will examine age assurance methods, default account settings, reporting tools, and the spread of illegal content. Regulators argue that self-declared age may be insufficient, while default settings and recommendations may expose minors to risks.

The Commission will now gather further evidence through information requests, inspections, and interviews, and may take enforcement actions, including interim measures or penalties.

National regulators will support the investigation as part of coordinated oversight under the Digital Services Act.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

New AI safety policies target teen protection in apps

OpenAI has released a set of prompt-based safety policies to help developers build safer AI experiences for teenagers. The tools work with the open-weight model gpt-oss-safeguard, turning safety requirements into practical classifiers for real-world use.

The policies address teen risks, including graphic violence, sexual content, harmful body image behaviour, dangerous challenges, roleplay, and age-restricted goods and services. Developers can use them for both real-time filtering and offline content analysis.

The framework was developed with input from organisations such as Common Sense Media and everyone.ai to improve clarity and consistency in teen safety rules. The initiative also responds to long-standing challenges in translating high-level safety goals into precise operational systems.

Open-source availability through the ROOST Model Community allows developers to adapt and expand the policies for different use cases and languages. The framework is a foundational step, not a complete solution, encouraging layered safeguards and ongoing refinement.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

UK tests social media bans for children in national pilot

The UK government has launched a large-scale pilot programme to test social media restrictions in the homes of 300 teenagers, aiming to improve children’s well-being instead of relying solely on existing digital safety measures.

The initiative, led by the Department for Science, Innovation and Technology and supported by Liz Kendall, will run for six weeks and examine how limits on digital platforms affect young people’s daily lives, including sleep, schoolwork, and family relationships.

Families across the UK will be divided into groups testing different approaches. Some parents will block access to social media entirely, while others will introduce a one-hour daily limit on popular platforms such as Instagram, TikTok, and Snapchat.

Another group will implement overnight curfews, restricting access between 9 pm and 7 am, while a control group will maintain existing usage patterns rather than introducing changes.

Participants will be interviewed before and after the trial to assess behavioural and practical outcomes, including how easily restrictions can be enforced and whether teenagers attempt to bypass controls.

The pilot runs alongside a national consultation on children’s digital well-being, which has already received nearly 30,000 responses. Government officials and academic experts will analyse data gathered from both initiatives to guide future policy decisions.

A programme that aims to ensure that any regulatory steps are evidence-based, reflecting real-life experiences rather than theoretical assumptions about digital behaviour.

Alongside the government trials, an independent scientific study funded by the Wellcome Trust will examine the effects of reduced social media use among adolescents.

Led by researchers from the University of Cambridge and the Bradford Institute for Health Research, the study will involve around 4,000 students aged 12 to 15.

Findings are expected to provide deeper insight into how social media influences anxiety, sleep, relationships, and overall well-being, supporting policymakers in shaping future online safety measures instead of relying on limited evidence.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

New UK rules target foreign influence and crypto donations

The UK government has announced sweeping reforms to political donations, introducing a £100,000 annual cap on contributions from overseas electors. The move targets concerns that individuals living abroad could exert disproportionate financial influence on domestic politics.

Cryptocurrency donations have also been banned with immediate effect, reflecting fears over anonymity and the difficulty of tracing funds. Authorities warn that digital assets risk enabling untraceable political funding until stronger regulation is in place.

Both measures will apply retrospectively, requiring political parties and candidates to return any unlawful donations within 30 days once the legislation takes effect. Enforcement action may follow for non-compliance, signalling a stricter approach to financial oversight.

Reforms stem from the Rycroft Review, which highlighted vulnerabilities in the UK’s electoral system linked to foreign interference. Further changes, including stronger Electoral Commission powers and tighter donor checks, are expected.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

ICO and Ofcom issue guidance on age assurance and online safety

The Information Commissioner’s Office and Ofcom have issued a joint statement outlining how age assurance measures should align with online safety and data protection requirements.

A guidance that focuses on protecting children from harm online instead of treating safety and privacy as separate obligations, reflecting closer coordination between the two regulators.

The statement is directed at digital services likely to be accessed by children and falling within the scope of the Online Safety Act and UK data protection laws.

It provides a practical overview of existing policies, helping organisations understand how to meet both regulatory frameworks while implementing age assurance technologies.

Rather than introducing new rules, the guidance clarifies how current requirements interact in practice. It highlights the importance of designing systems that both verify users’ ages and safeguard personal data, ensuring that safety measures do not undermine privacy protections.

The approach encourages organisations to integrate compliance into service design instead of addressing obligations separately.

By aligning regulatory expectations, the ICO and Ofcom aim to support organisations in delivering safer online environments for children while maintaining strong data protection standards.

The joint effort signals a broader move towards coordinated digital regulation, where safety and privacy are addressed together to reflect the complexities of modern online services.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Digital Services Act disinformation signatories publish first 2026 reports

Signatories to the EU Code of Conduct on Disinformation have published new transparency reports describing the measures they say they are taking to reduce the spread of disinformation online. According to the European Commission, the reports are the first ones submitted since the Code was recognised as a code of conduct under the Digital Services Act.

The reports are available through the Code’s Transparency Centre and come from a broad group of signatories, including online platforms such as Google, Meta, Microsoft, and TikTok, as well as fact-checkers, research organisations, civil society bodies, and representatives of the advertising industry. The European Commission says the reporting round covers the period from 1 July to 31 December 2025 and marks the first full reporting cycle linked to the Digital Services Act.

Dedicated sections in the reports cover responses to ongoing crises, notably the conflict in Ukraine, as well as measures intended to safeguard the integrity of elections. Data on the implementation of disinformation-related measures is also included, alongside developments in signatories’ policies, tools, and partnerships under the Digital Services Act framework.

Greater significance attaches to the reporting cycle because of the Code’s changed legal and regulatory position. The Commission says the Code was endorsed on 13 February 2025 by the Commission and the European Board for Digital Services, at the request of the signatories, as a code of conduct within the meaning of the Digital Services Act. From 1 July 2025, the Code became part of the co-regulatory framework under the Digital Services Act.

A more formal role now applies to the Code than under its earlier voluntary setup. According to the Commission, signatories’ adherence to its commitments is subject to independent annual auditing, and the Code serves as a relevant benchmark for determining compliance with Article 35 of the Digital Services Act. The Commission also says the Code has become a ‘significant and meaningful benchmark of DSA compliance’ for providers of very large online platforms and very large online search engines that adhere to its commitments under the Digital Services Act.

Reporting obligations differ depending on the type of signatory. Under the Code, providers of very large online platforms and very large online search engines commit to reporting, every six months, on the actions taken by their subscribed services. The Commission lists Google Search, YouTube, Google Ads, Facebook, Instagram, Messenger, WhatsApp, Bing, LinkedIn, and TikTok among the covered services, while other non-platform signatories report once per year under the Digital Services Act structure.

Broader policy relevance lies in the EU’s attempt to connect platform self-reporting to a more formal oversight structure. By placing the disinformation Code inside the Digital Services Act framework, the Commission and the Board are using voluntary commitments, transparency reporting, and auditing as part of a co-regulatory approach to systemic online risks. The reports themselves do not prove compliance, but they now carry greater weight within the wider Digital Services Act architecture for platform governance.

One further point is that the Commission notice focuses on publication of the reports rather than evaluating their quality or effectiveness. The notice says the reports describe measures, data, and policy developments, but it does not assess whether the actions taken by signatories were sufficient. Such a distinction matters in politically sensitive areas such as election integrity and crisis-related disinformation, especially where transparency under the Digital Services Act may shape future scrutiny.

Taken together, the first reporting round shows how the EU is using the Digital Services Act not only to impose direct legal obligations on large platforms and search engines, but also to anchor voluntary commitments within a more structured regulatory environment. Continued reporting, auditing, and review will determine how much practical weight the Code carries within the Digital Services Act and how effectively the Digital Services Act supports oversight of disinformation risks online.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!

Australian regulator warns AI companions expose children to serious online risks

The eSafety Commissioner has reported that AI companion chatbots are failing to adequately protect children from harmful content, following a transparency review of services including Character.AI, Nomi, Chai, and Chub AI.

According to the report, these services did not implement robust safeguards against exposure to sexually explicit material or the generation of child sexual exploitation and abuse content.

The findings also indicate that most platforms relied on self-declared age verification and did not consistently monitor inputs or outputs across all AI models used.

eSafety Commissioner Julie Inman Grant stated that AI companions, often presented as sources of emotional or social support, are increasingly used by children but may expose them to harmful interactions.

She noted that none of the reviewed services had ‘meaningful age checks’ in place and highlighted concerns about the absence of safeguards related to self-harm and suicide content.

The report further identifies that several platforms in Australia did not refer users to crisis or mental health support services when harmful interactions were detected.

It also notes gaps in monitoring for unlawful content and limited investment in trust and safety staffing, with some providers reporting no dedicated moderation personnel.

The findings follow the implementation of Australia’s Age-Restricted Material Codes, which require online services, including AI chatbots, to prevent access to age-inappropriate content and provide appropriate safety measures.

These obligations complement existing Unlawful Material Codes and Standards, with non-compliance potentially leading to civil penalties.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!