IF / AI / 2026
AI news, new models, science and technology already changing your phone, your work and decisions made by companies and governments.
AI HISTORY / SEVEN MOMENTS
Pick a year to read what happened, why it mattered and what changed afterwards.
Models now browse, research, handle files, write code and coordinate multi-step work. 'Agent' is the fashionable word, even if every company uses it slightly differently.
When agents work, they can save hours. When they make a mistake after completing twenty steps and still deliver a persuasive result, finding the error becomes much harder than correcting one wrong chat reply.
Labs are competing to build systems that can finish useful work without supervision at every click.
CURRENT ARCHIVE
01OPENFrom August 2, chatbots must identify themselves at the start and deepfakes need a visible label. Publishers keep an exemption when a person reviews the text and takes responsibility for it.
02OPENMaking a song has never been easier. Deezer now receives nearly 75,000 fully AI-generated tracks a day. Uploading them is cheap; giving anyone a reason to listen—and remember who made them—is the harder part.
03OPENKathy Hochul has frozen state permits for new data centers of 50MW or more for up to a year. New York wants to know how much power and water they will use—and, above all, who will pay for the infrastructure they need.
04OPENAlpha Chat already works without an account, shows what it uses to answer and deletes conversations after 30 days. It is only version 0.1, but the coalition behind it has more than $400 million pledged and wants to turn open AI into a public service.
05OPENSan Francisco’s city attorney has identified eight iOS apps and five Android apps that could turn ordinary photographs into non-consensual sexual images. Google says it removed all five flagged Play Store apps; Apple has removed three and warned four more developers. European rules will soon make effective safeguards a condition of selling these tools across the EU.
06OPENQwen-AgentWorld simulates terminals, websites, search, Android and MCP tools so other agents can practise without touching live systems. Its training results are promising, although the benchmark placing it above GPT-5.4 needs important context.
07OPENChina's 2.8-trillion-parameter model can handle a million-token context and has impressed coding evaluators. It is already reviving DeepSeek-era anxiety in the US, even though the weights, licence and technical report have not arrived.
08OPENClaude for Teachers can draft lessons, adapt worksheets and review class data while educators head home. The free year is generous; the contest to become part of school life is much bigger.
09OPENOpenAI is launching a model built to research, use tools and complete long assignments. It could save hours, but tracing a mistake across all those steps will be harder.
10OPENDeepMind and Isomorphic Labs want to speed up biological-threat detection and treatment design. But models that help defend us may also make harm cheaper.
11OPENOpenAI, Anthropic and Google DeepMind increasingly agree on federal rules. The decisive question is who writes them, how costly they are and which competitors can comply.