🏫 In some schools, chatbots interrogate students about their work. But the AI revolution has teachers worried ↗
Some Australian schools are trialling a sharper twist on “show your working” - students finish an assignment, then an AI chatbot quizzes them in a back-and-forth to check they understand what they submitted. It’s basically an oral viva, but with a machine asking “what do you mean by that?” until you start to squirm.
The bigger worry is uneven rollout: an industry paper warns that fast adopters could sprint ahead while others lag, creating a two-speed education system. Teachers also keep flagging the familiar - and still serious - concerns: privacy, wellbeing, and how to stop “AI help” sliding into “AI did it”.
👮 Met police using AI tools supplied by Palantir to flag officer misconduct ↗
London’s Met is using Palantir-supplied AI to analyse internal workforce patterns - sickness, absences, overtime - looking for signals that might correlate with professional standards issues. It’s pitched as a time-limited pilot and, crucially, the Met says humans still make the actual determinations… but that “still” is doing a lot of work.
The Police Federation isn’t thrilled, calling it “automated suspicion” and warning about opaque tools misreading stress and workload as wrongdoing. It’s one of those moments where the tech sounds neat on a slide, then you picture how it feels on the ground and it turns sharply prickly.
🛡️ UK privacy watchdog warns over AI-generated images in joint statement ↗
The UK’s ICO joined other regulators in a joint warning about AI-generated images that depict identifiable people without consent - basically: don’t treat “it’s synthetic” as a magic excuse. The message is to build safeguards early and engage regulators instead of doing the classic tech move of shipping first and apologising later.
They also highlight heightened risks for children, which makes sense, because once this stuff spreads it’s like trying to scoop glitter back into a jar. The thrust is that privacy and dignity are meant to survive the fashion cycle, or so it seems.
🪖 US Defense Secretary Hegseth summons Anthropic CEO for tough talks over military use of Claude, Axios reports ↗
Reuters reports that the U.S. Defense Secretary is set to meet Anthropic’s CEO for what’s described as a more confrontational conversation about military use of Claude. The Pentagon reportedly wants major AI providers to make models available on classified networks with fewer restrictions - and Anthropic has been resisting any loosening of safeguards.
The subtext reads like a classic “move fast but… with missiles involved” dilemma. If the parties can’t agree on guardrails, the relationship could strain or even snap - yet Anthropic is also signalling that talks with government remain constructive, so it’s tense, but not full break-up-texts (yet).
🧠 ChatGPT - Release Notes ↗
OpenAI’s release notes point to an expanded context window for “Thinking” mode - meaning you can feed the model more text and keep longer threads coherent without it dropping earlier details. If you’ve ever watched a model “forget” something you literally just said (pain), this is aimed right at that.
It’s a very nerdy upgrade that quietly changes what’s practical: longer docs, more tangled projects, more sprawling research sessions. Not glamorous, but it’s the kind of plumbing change that suddenly makes the whole house feel bigger.
FAQ
What does it mean when schools use chatbots to “interrogate” students about assignments?
In some Australian schools, students complete an assignment and then go through a back-and-forth with an AI chatbot that probes their reasoning. It works like an oral viva, pressing students to explain what they wrote and why. The purpose is to verify understanding, not just output. It can also expose gaps when a student cannot defend key claims.
How could AI chatbots in schools create a “two-speed” education system?
An industry paper warns that fast adopters could pull ahead while slower-moving schools slip further back. If some schools deploy AI tools to assess understanding, support learning, or deter misuse, they may lift outcomes sooner. Others may lack budget, staff capacity, or policy clarity to roll out comparable systems. That uneven uptake can widen existing resource and attainment gaps.
How are teachers trying to stop “AI help” turning into “AI did it”?
A common approach is to shift assessment toward process and explanation, not just the finished product. Chatbot-style questioning, drafts, reflections, or oral checks can make it harder to submit work you do not understand. Teachers may also set firmer boundaries on acceptable assistance, such as brainstorming versus full rewriting. The emphasis stays on demonstrating understanding, not policing every tool.
What privacy and wellbeing issues come up when AI is used in classrooms?
Teachers keep raising concerns about what student data is collected, where it is stored, and who can access it. There are also wellbeing worries if students feel perpetually monitored or pressured by machine-led questioning. Children are a special risk group, because sensitive information can spread or persist in systems in unexpected ways. Many schools aim to put safeguards in place early rather than trying to retrofit them later.
How is the Met using Palantir-supplied AI to flag potential officer misconduct?
London’s Met is using Palantir-supplied AI to analyse workforce patterns such as sickness, absences, and overtime. The aim is to spot signals that might correlate with professional standards issues. It is described as a time-limited pilot, and the Met says humans still make the determinations. Critics argue the tool can feel like “automated suspicion” if it reads stress as wrongdoing.
Why are police unions worried about “automated suspicion” from AI oversight tools?
The Police Federation’s concern is that opaque models may infer misconduct from patterns that have benign explanations. Workload spikes, health issues, and stress can shift attendance or overtime without implying wrongdoing. If the tool’s logic is not transparent, it becomes harder for officers to challenge flags or for managers to calibrate fair thresholds. Even with human decisions, automated signals can still shape outcomes.
What is the UK ICO warning about with AI-generated images of identifiable people?
The UK privacy watchdog joined other regulators in warning that “synthetic” does not mean consequence-free. If an AI-generated image depicts an identifiable person without consent, privacy and dignity risks still apply. Regulators stress building safeguards early and engaging with oversight rather than shipping first. They also flag heightened risks for children, where harm can scale quickly once content spreads.
What’s behind the reported Pentagon pressure on Anthropic about military use of Claude?
Reuters reports the U.S. Defense Secretary is set to meet Anthropic’s CEO for tougher talks on military use of Claude. The Pentagon reportedly wants major AI providers to make models available on classified networks with fewer restrictions. Anthropic has been resisting looser safeguards, pointing to a clash over guardrails. The relationship sounds tense, though both sides signal talks remain constructive.
What changed in the ChatGPT release notes about “Thinking” mode context length?
OpenAI’s release notes point to an expanded context window for “Thinking” mode, allowing more text to sit within a single thread. In practice, that can help with long documents, complex projects, and multi-step research where earlier details matter. It is a plumbing-style upgrade rather than a flashy feature. The main benefit is fewer dropped details over longer sessions.