Artificial intelligence is increasingly capable of handling everyday tasks, yet who benefits depends heavily on where people live, what language they speak and how much control they have over the technology. Recent developments discussed on This Week in AI examined advances from OpenAI and Anthropic, the rise of regional models and the practical challenges of making AI both accessible and responsible.
Frontier AI entering everyday services
OpenAI introduced LifeVal, a benchmark designed to measure how well AI handles routine tasks such as finding childcare, managing household budgets and accessing government services. According to OpenAI, LifeVal scores rose from 49% for GPT-5.2 to roughly 80% for GPT-5.6 and GPT-6 Astra. The benchmark is intended to help researchers assess whether capability gains translate into useful assistance for daily responsibilities.
OpenAI also reported that ChatGPT now handles roughly 20% of Google’s search volume in search-like queries, though the company has not disclosed its methodology. The firm made a GPT-6 Cyber model available to approved security teams to help identify and fix software vulnerabilities.
Anthropic has begun testing read-only banking connections in Claude for some U.S. subscribers, enabling users to analyze spending and flag unusual charges. As AI assistants take on tasks previously performed by banks, search engines and specialized software, those industries must rethink how they serve customers and protect sensitive data.
Anthropic also published FlashProve, a research prototype that uses cryptographic proofs to verify which model produced a response. Separately, the nonprofit research organization METR reported that AI assistance may have roughly doubled the pace of progress on certain model development tasks at Anthropic over the past year. While the findings do not demonstrate recursive self-improvement, they suggest AI help is already accelerating parts of model development, increasing the need to validate outputs and understand how AI shapes progress.
Regional AI to close access gaps
Previous discussions have addressed the “tokenizer tax,” where some languages cost more to process because they require more tokens than English. New research from the University of Michigan identified an additional disparity: in one test, answering questions in Pashto consumed up to 179 times more energy than answering equivalent questions in English, while accuracy also suffered. The result illustrates how language affects both the cost and quality of AI services.
Countries are responding by investing in models tailored to their languages and infrastructure. Ibero-American countries agreed to form a regional AI alliance to coordinate model development, data sharing and compute resources. India introduced three domestic models, including EKA-1, which supports 22 Indian languages and English, and two Dhīr models designed to work offline with documents. Offline capabilities could be particularly valuable in rural areas with unreliable internet access.
Other local projects include Indonesia’s GoTo, the UAE’s Jais 3, and Uganda’s Sunflower V2. These initiatives may give developers more control over AI for local languages and reduce dependence on proprietary services, even when some systems build on open foundation models from larger labs. For organizations deploying AI across multiple countries, testing performance and cost in each language can reveal problems that English-language benchmarks overlook.
Responsible AI depends on deployment choices
Policy and legal disputes are shaping how AI is used. Google faced criticism after changing AI controls for school accounts: under a new arrangement, schools that disable AI also lose access to Google Search, limiting educators’ control over which tools students can use. In Europe, an advisor to the EU’s top court suggested chatbots might infringe publishers’ rights when reproducing news content; the opinion arose from a dispute involving Google’s Gemini and a Hungarian publisher. A final ruling is pending, but the case could affect how AI firms access and reuse copyrighted material. This is part of a broader wave of publisher lawsuits, including The New York Times’ U.S. case against OpenAI and Microsoft alleging use of its articles to train models and reproduce content in ChatGPT outputs.
Environmental costs and local opposition
The environmental costs of AI infrastructure are receiving attention. Some data center operators are repurposing retired jet engines as natural-gas generators to obtain power while awaiting grid connections. Although this can speed deployment, converted engines are less efficient and can emit more pollution than conventional turbines, trading faster access to electricity for higher emissions and worse local air quality. Public opposition is rising: in the November 3 U.S. general elections, voters favored restrictions on data center development in 40 of 41 local ballot measures across 10 states.
At the same time, Planet introduced a natural-language tool for searching and analyzing more than a decade of satellite imagery. The application lets users explore geographic changes without specialized GIS expertise, demonstrating how AI can broaden access to complex data that previously required specialized tools and training.
What’s next
As AI expands into personal services and countries invest in their own models, developers and organizations face increasingly diverse requirements for security, performance and accessibility. The most useful systems will be those that work reliably for the people and circumstances they are designed to serve.
This Week in AI will return on Monday, November 16, with another episode covering the latest developments in AI research and applications. Previous episodes are available on YouTube, Spotify, Apple and other podcast platforms.
