The German government has unveiled a national AI strategy aimed at making artificial intelligence a key economic driver by 2030. Dubbed an “AI offensive,” the plan targets 10% of Germany’s GDP to be AI-based by 2030, seeking to narrow the technology gap with the US and China. The draft strategy outlines investments in large-scale AI processing centers (“AI Gigafactories”) coordinated with industry and federal states. Brussels is backing infrastructure expansion with €20 billion earmarked by the European Commission for AI/data facilities. The German plan also sets goals in quantum computing (two error-corrected quantum computers by 2030) and emphasizes public-private cooperation. The Cabinet is expected to approve the strategy within the month, underscoring Germany’s commitment to AI-driven innovation.

The European Commission has published a General-Purpose AI Code of Practice to help providers of large AI models align with upcoming EU regulations. Drafted by 13 experts with input from over 1,000 stakeholders, the voluntary Code offers guidelines on transparency, copyright, safety, and security for foundation-model AI. It supports compliance with the EU’s AI Act rules taking effect in August 2025, which will require makers of broadly-used AI systems to implement risk assessments and transparency measures. Companies that sign the Code and follow its recommended practices will enjoy greater legal certainty and reduced bureaucracy. This initiative reflects Europe’s push for responsible AI innovation, ensuring powerful AI models are safe, transparent, and trustworthy.

At UNESCO’s 3rd Global Forum on AI Ethics in Bangkok, international leaders called for turning AI ethics principles into concrete actions. The forum highlighted progress since UNESCO’s landmark 2021 AI Ethics Recommendation, noting that dozens of countries have begun integrating ethical frameworks and conducting AI readiness assessments. New global cooperation platforms were announced, including a Network of AI Supervisory Authorities and a civil society & academia network, to strengthen cross-border governance of AI. A key message was that international and regional collaboration is urgently needed to ensure AI develops in a human-centric, responsible manner. Officials stressed moving beyond principles: implementing oversight mechanisms and sharing best practices so that ethical guidelines lead to real-world impact.

A new annual State of AI report reveals that the global AI landscape is more competitive than ever. According to the analysis (from Stanford’s HAI), high-performing Chinese AI models have sharply increased in number and quality, rapidly challenging US leadership in the field. The performance edge between top American and Chinese models is now narrowing, as evidenced by benchmark tests and research output. Additionally, 2024 saw breakthrough progress in efficient AI systems, with smaller, optimized models matching the capabilities of previous large “behemoth” models. These trends indicate a tightening AI race, driven by China’s heavy R&D investments and talent growth. Experts say businesses worldwide should watch this evolving balance, as it may spur faster innovation and more diverse AI solutions.

At its Build 2025 conference, Microsoft announced new tools for organizations to tailor AI to their needs. The company introduced Microsoft 365 Copilot “Tuning”, a low-code service that lets enterprises train Microsoft’s AI Copilot on their own data, workflows, and style guides. This enables companies to create domain-specific AI assistants – for example, a law firm can generate an agent that drafts documents in the firm’s preferred format and tone. Microsoft also unveiled multi-agent orchestration in Copilot Studio, allowing multiple specialized agents to collaborate on complex tasks. These updates, delivered within Microsoft’s secure cloud boundary, let businesses harness AI more accurately and securely.

IBM is doubling down on enterprise AI with new integration tools and a major investment pledge. CEO Arvind Krishna outlined a strategy to help clients orchestrate fleets of AI “agents” – software bots from IBM and third-parties – across business applications. Unveiled at the annual IBM Think conference, the approach allows companies to plug in AI solutions from Salesforce, Workday, Adobe, and others into IBM’s platform, while quickly building their own agents using IBM’s Granite AI models. The company also announced it will invest $150 billion in the US over five years to strengthen its manufacturing and R&D in mainframes, artificial intelligence, and quantum computing.

The European Commission has published new guidelines on the responsible use of generative AI in research, aiming to foster ethics and transparency in scientific innovation. The recommendations emphasize four core principles for researchers: Reliability, Honesty, Respect, and Accountability. This means AI models used in research should be robust and unbiased, methods and data handling should be transparent and fair, and privacy, intellectual property, and human dignity must be safeguarded. Researchers are urged to clearly label AI-generated content and ensure human oversight in study design. These guidelines, while non-binding, are intended to shape responsible AI innovation in academic and applied fields.

Google has revealed Ironwood, its seventh-generation Tensor Processing Unit (TPU) custom chip, designed to power the next wave of AI applications. Announced at the Google Cloud Next ’25 event, Ironwood is purpose-built for running advanced AI inference (“the age of inference”) with unprecedented scale and efficiency. A full Ironwood pod links 9,216 chips together to provide 42.5 exaflops of compute – over 24 times more processing power than the world’s largest supercomputer today. Google says Ironwood offers a 10× performance boost over its previous TPUs by improving energy efficiency and memory bandwidth.

Google has made its latest Gemini 2.5 Pro AI model – known for enhanced “reasoning” capabilities – freely available for anyone to try. This comes just days after the model’s preview release to paying subscribers. The Gemini 2.5 Pro incorporates step-by-step logical reasoning (“chain-of-thought”) that significantly improves performance on complex tasks. Google reports the model now leads on several language-model benchmarks. While usage is currently rate-limited, opening access widely marks a shift toward more inclusive AI development. Google also plans to integrate this advanced model into its mobile Gemini app and developer tools.

The European Commission has released an updated set of Clauses for AI procurement (MCC-AI), giving public agencies a template to demand responsible AI practices from vendors. The clauses align with the EU’s draft AI Act, mirroring obligations for “high-risk” AI systems (e.g. in hiring or healthcare). They cover key areas such as risk assessment, data governance, transparency, human oversight, and cybersecurity measures. The MCC-AI package includes a full version for high-risk AI, a lighter version for lower-risk cases, and detailed guidance. While designed for public-sector contracts, these standards can be adopted by companies as well to foster compliance and trust.

Google has updated its AI ethics principles, removing earlier explicit commitments to avoid certain high-risk applications. The new version emphasizes broader terms like “responsible AI” and “collaboration,” omitting previous categorical restrictions. According to Google, this shift is necessary to remain competitive and to support democratic values in the evolving AI ecosystem. Critics, however, argue that the changes weaken the company’s previous stance on harm prevention and transparency. The update reignites industry-wide debate on how ethical frameworks should be implemented and maintained.

On February 2, 2025, the EU AI Act officially begins enforcement of key restrictions on AI applications considered to pose an “unacceptable risk.” Banned systems include those that manipulate user behavior or implement harmful social scoring. Biometric surveillance in public is also prohibited. This regulation aims to preserve fundamental rights such as privacy and equality and sets a strong precedent for global AI governance. Companies that violate these rules face fines of up to 7% of global turnover.

OpenAI is planning a new $40 billion funding round, with SoftBank potentially contributing $25 billion. The funding would support a $500 billion initiative to build a next-generation AI infrastructure platform called “Stargate.” This project, aimed at scaling compute power for future AI models, highlights the intensifying race to build global AI capabilities. If finalized, the deal could mark one of the most significant AI infrastructure investments to date.

OpenAI has released “o3-mini,” a streamlined AI model optimized for transparency and logic-based reasoning. It processes tasks about 24% faster than previous versions and includes visible step-by-step outputs. The model is available to free ChatGPT users with usage limits and can also be accessed via API. This release represents a shift toward more transparent, efficient AI models for both consumer and enterprise use cases.

Ireland’s data regulator has fined Meta €251 million following a long-standing privacy breach involving Facebook. The 2018 issue exposed personal data of 29 million users, including contact details and locations. Meta was found to have insufficiently implemented data protection measures and failed to report the breach transparently. This decision reinforces the EU’s strict enforcement of data compliance rules under GDPR.

Researchers at MIT have introduced a new method to reduce bias in machine learning models. The approach down-weights specific training data points that reduce performance for underrepresented groups, without degrading overall model accuracy. This method offers a targeted, scalable alternative to conventional dataset rebalancing and shows promise for enhancing fairness in real-world AI deployments.