Sustainable development Archives | Page 49 of 285

AI firms fall short of EU transparency rules on training data

Several major AI companies appear slow to meet EU transparency obligations, raising concerns over compliance with the AI Act.

Under the regulation, developers of large foundation models must disclose information about training data sources, allowing creators to assess whether copyrighted material has been used.

Such disclosures are intended to offer a minimal baseline of transparency, covering the use of public datasets, licensed material and scraped websites.

While open-source providers such as Hugging Face have already published detailed templates, leading commercial developers have so far provided only broad descriptions of data usage instead of specific sources.

Formal enforcement of the rules will not begin until later in the year, extending a grace period for companies that released models after August 2025.

The European Commission has indicated willingness to impose fines if necessary, although it continues to assess whether newer models fall under immediate obligations.

The issue is likely to become politically sensitive, as stricter enforcement could affect US-based technology firms and intensify transatlantic tensions over digital regulation.

Transparency under the AI Act may therefore test both regulatory resolve and international relations as implementation moves closer.

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

Anthropic report shows AI is reshaping work instead of replacing jobs

A new report by Anthropic suggests fears that AI will replace jobs remain overstated, with current use showing AI supporting workers rather than eliminating roles.

Analysis of millions of anonymised conversations with the Claude assistant indicates technology is mainly used to assist with specific tasks rather than full job automation.

The research shows AI affects occupations unevenly, reshaping work depending on role and skill level. Higher-skilled tasks, particularly in software development, dominate use, while some roles automate simpler activities rather than core responsibilities.

Productivity gains remain limited when tasks grow more complex, as reliability declines and human correction becomes necessary.

Geographic differences also shape adoption. Wealthier countries tend to use AI more frequently for work and personal activities, while lower-income economies rely more heavily on AI for education. Such patterns reflect different stages of adoption instead of a uniform global transformation.

Anthropic argues that understanding how AI is used matters as much as measuring adoption rates. The report suggests future economic impact will depend on experimentation, regulation and the balance between automation and collaboration, rather than widespread job displacement.

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

South Korea faces mounting pressure from US AI chip tariffs

New US tariffs on advanced AI chips are drawing scrutiny over their impact on global supply chains, with South Korea monitoring potential effects on its semiconductor industry.

The US administration has approved a 25 percent tariff on advanced chips that are imported into the US and then re-exported to third countries. The measure is widely seen as aimed at restricting the flow of AI accelerators to China.

The tariff thresholds are expected to cover processors such as Nvidia’s H200 and AMD’s MI325X, which rely on high-bandwidth memory supplied by Samsung Electronics and SK hynix.

Industry officials say most memory exports from South Korea to the US are used in domestic data centres, which are exempt under the proclamation, reducing direct exposure for suppliers.

South Korea’s trade ministry has launched consultations with industry leaders and US counterparts to assess risks and ensure Korean firms receive equal treatment to competitors in Taiwan, Japan and the EU.

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

Energy-efficient AI training with memristors

Scientists in China developed an error-aware probabilistic update (EaPU) to improve neural network training on memristor hardware. The method tackles accuracy and stability limits in analog computing.

Training inefficiency caused by noisy weight updates has slowed progress beyond inference tasks. EaPU applies probabilistic, threshold-based updates that preserve learning and sharply reduce write operations.

Experiments and simulations show major gains in energy efficiency, accuracy and device lifespan across vision models. Results suggest broader potential for sustainable AI training using emerging memory technologies.

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

AI power demand pushes nuclear energy back into focus

Rising AI-driven electricity demand is straining power grids and renewing focus on nuclear energy as a stable, low-carbon solution. Data centres powering AI systems already consume electricity at the scale of small cities, and demand is accelerating rapidly.

Global electricity consumption could rise by more than 10,000 terawatt-hours by 2035, largely driven by AI workloads. In advanced economies, data centres are expected to drive over a fifth of electricity-demand growth by 2030, outpacing many traditional industries.

Nuclear energy is increasingly positioned as a reliable backbone for this expansion, offering continuous power, high energy density, and grid stability.

Governments, technology firms, and nuclear operators are advancing new reactor projects, while long-term power agreements between tech companies and nuclear plants are becoming more common.

Alongside large reactors, interest is growing in small modular reactors designed for faster deployment near data centres. Supporters say these systems could ease grid bottlenecks and deliver dedicated power for AI, strengthening nuclear energy’s role in the digital economy.

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

xAI faces stricter pollution rules for Memphis data centre

US regulators have closed a loophole that allowed Elon Musk’s AI company, xAI, to operate gas-burning turbines at its Memphis data centre without full air pollution permits. The move follows concerns over emissions and local health impacts.

The US Environmental Protection Agency clarified that mobile gas turbines cannot be classified as ‘non-road engines’ to avoid Clean Air Act requirements. Companies must now obtain permits if their combined emissions exceed regulatory thresholds.

Local authorities had previously allowed the turbines to operate without public consultation or environmental review. The updated federal rule may slow xAI’s expansion plans in the Memphis area.

The Colossus data centre, opened in 2024, supports training and inference for Grok AI models and other services linked to Musk’s X platform. NVIDIA hardware is used extensively at the site.

Residents and environmental groups have raised concerns about air quality, particularly in nearby communities. Legal advocates say xAI’s future operations will be closely monitored for regulatory compliance.

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

EU revises Cybersecurity Act to streamline certification

The European Commission plans to revise the Cybersecurity Act to expand certification schemes beyond ICT products and services. Future assessments would also cover companies’ overall risk-management posture, including governance and supply-chain practices.

Only one EU-wide scheme, the Common Criteria framework, has been formally adopted since 2019. Cloud, 5G, and digital identity certifications remain stalled due to procedural complexity and limited transparency under the current Cybersecurity Act framework.

The reforms aim to introduce clearer rules and a rolling work programme to support long-term planning. Managed security services, including incident response and penetration testing, would become eligible for EU certification.

ENISA would take on a stronger role as the central technical coordinator across member states. Additional funding and staff would be required to support its expanding mandate under the newer cybersecurity laws.

Stakeholders broadly support harmonisation to reduce administrative burden and regulatory fragmentation. The European Commission says organisational certification would assess cybersecurity maturity alongside technical product compliance.

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

OpenAI outlines advertising plans for ChatGPT access

The US AI firm, OpenAI, has announced plans to test advertising within ChatGPT as part of a broader effort to widen access to advanced AI tools.

An initiative that focuses on supporting the free version and the low-cost ChatGPT Go subscription, while paid tiers such as Plus, Pro, Business, and Enterprise will continue without advertisements.

According to the company, advertisements will remain clearly separated from ChatGPT responses and will never influence the answers users receive.

Responses will continue to be optimised for usefulness instead of commercial outcomes, with OpenAI emphasising that trust and perceived neutrality remain central to the product’s value.

User privacy forms a core pillar of the approach. Conversations will stay private, data will not be sold to advertisers, and users will retain the ability to disable ad personalisation or remove advertising-related data at any time.

During early trials, ads will not appear for accounts linked to users under 18, nor within sensitive or regulated areas such as health, mental wellbeing, or politics.

OpenAI describes advertising as a complementary revenue stream rather than a replacement for subscriptions.

The company argues that a diversified model can help keep advanced intelligence accessible to a wider population, while maintaining long term incentives aligned with user trust and product quality.

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

New Steam rules redefine when AI use must be disclosed

Steam has clarified its position on AI in video games by updating the disclosure rules developers must follow when publishing titles on the platform.

The revision arrives after months of industry debate over whether generative AI usage should be publicly declared, particularly as storefronts face growing pressure to balance transparency with practical development realities.

Under the updated policy, disclosure requirements apply exclusively to AI-generated material consumed by players.

Artwork, audio, localisation, narrative elements, marketing assets and content visible on a game’s Steam page fall within scope, while AI tools used purely during development remain outside Valve’s interest.

Developers using code assistants, concept ideation tools or AI-enabled software features without integrating outputs into the final player experience no longer need to declare such usage.

Valve’s clarification signals a more nuanced stance than earlier guidance introduced in 2024, which drew criticism for failing to reflect how AI tools are used in modern workflows.

By formally separating player-facing content from internal efficiency tools, Steam acknowledges common industry practices without expanding disclosure obligations unnecessarily.

The update offers reassurance to developers concerned about stigma surrounding AI labels while preserving transparency for consumers.

Although enforcement may remain largely procedural, the written clarification establishes clearer expectations and reduces uncertainty as generative technologies continue to shape game production.

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

New ETSI standard defines cybersecurity rules for AI systems

ETSI has released ETSI EN 304 223, a new European Standard establishing baseline cybersecurity requirements for AI systems.

Approved by national standards bodies, the framework becomes the first globally applicable EN focused specifically on securing AI, extending its relevance beyond European markets.

The standard recognises that AI introduces security risks not found in traditional software. Threats such as data poisoning, indirect prompt injection and vulnerabilities linked to complex data management demand tailored defences instead of conventional approaches alone.

ETSI EN 304 223 combines established cybersecurity practices with targeted measures designed for the distinctive characteristics of AI models and systems.

Adopting a full lifecycle perspective, the ETSI framework defines thirteen principles across secure design, development, deployment, maintenance and end of life.

Alignment with internationally recognised AI lifecycle models supports interoperability and consistent implementation across existing regulatory and technical ecosystems.

ETSI EN 304 223 is intended for organisations across the AI supply chain, including vendors, integrators and operators, and covers systems based on deep neural networks, including generative AI.

Further guidance is expected through ETSI TR 104 159, which will focus on generative AI risks such as deepfakes, misinformation, confidentiality concerns and intellectual property protection.

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