AI2Day Daily Brief — 29 Aug 2026

Good morning — it is Saturday the twenty-ninth of August, and we are opening with Meta quietly patching a privacy hole in its smart glasses after videos of strangers filmed through a hidden camera spread across the internet.

Meta is pushing a software update to its Ray-Ban Meta smart glasses after users found a straightforward workaround to its recording indicator light. The glasses have a small LED that is supposed to glow whenever the camera is active. The loophole worked like this: start recording first, then cover the light with tape or a sticker. The camera kept running with no visible warning to anyone nearby. Videos filmed that way circulated widely online. Meta's executive vice president of wearables, Alex Himel, announced the fix on Threads. The update will cut the recording if the light is covered mid-shoot. Meta has not said it addresses the broader question of covert filming in public spaces — this is a targeted technical patch, not a policy overhaul.

A precise fix for a precise exploit, though the wider consent questions are still on the table. Next, Apple's machine learning research team has published work on a tool that writes its own AI tests.

Testing an AI agent is genuinely difficult. These are systems that carry out multi-step tasks by calling external tools, checking results, and trying again across several conversation turns. Until now, building realistic test scenarios meant hiring domain experts to write them by hand — slow, expensive, and fragile the moment an external tool updates its behaviour. Apple M L Research's answer is a system called Agent Seer. It reads the specification of whichever tools an agent can use — the function names, descriptions, and expected inputs — and generates evaluation scenarios directly from that. No human authorship required, and no need to run live tools during the process, which cuts cost and complexity. The team says current benchmarks go stale as A P Is change; Agent Seer is designed to keep pace.

Automated test generation for automated systems — a compounding problem getting a compounding solution. Now to the warehouse floor, where Locus Robotics has bought a Canadian startup to solve one of automation's stubbornest physical problems.

Suction-cup grippers handle somewhere between sixty and seventy percent of warehouse items well. The rest — floppy polybags, crinkled fabric, awkwardly shaped containers — defeat them. Locus Robotics acquired Nexera Robotics, based in Vancouver, to get its NeuraGrasp technology: a soft gripper that combines suction and pinching together. Nexera had gone through at least six product generations before the acquisition. Locus's Array robot launched in March twenty twenty-five with a standard suction gripper; the company says its next public appearance will likely feature NeuraGrasp. Locus executives are framing pick quality, not speed, as the priority, pointing to product damage and missed picks as hidden costs that rarely surface in competitor marketing. Two broader industry problems remain open: A I decision-making that still lags behind the hardware, and touch sensors that wear out faster than anyone would like.

Sixty to seventy percent coverage sounds decent until you are the warehouse running the other thirty. Moving to Europe now, where Britain's government has rejected a Green party call to halt A I data centre construction.

Zack Polanski, co-leader of the Green party, called this year for a full moratorium on large A I data centres in Britain, describing them as, in his words, energy-guzzling, water-guzzling facilities that the country should slam the brakes on. His proposal would ban new large facilities while allowing smaller, community-scale centres to continue. Labour's government said flatly that such a moratorium would be, in ministers' words, a disaster for jobs and national security. The row puts two legitimate pressures in direct collision: the electricity and water demands of large-scale compute infrastructure on one side, and the economic and strategic case for building that infrastructure domestically on the other. Neither side is manufacturing the tension — it is a real trade-off with no clean answer.

Today's sponsor is Zero2Hero: an AI poker coach sits at your table, and you can ask it mid hand whether the player across from you is bluffing. The free tier is real: 50 hands a day against AI opponents, no card. Not a trial. Lifetime access is $99, in US dollars, until 15 September, at zero, the number two, hero, dot A I.

The same tension is playing out at national scale on the other side of the world. Australia's energy regulator has published a forecast that stops the conversation.

Australia's Energy Market Operator publishes an annual electricity supply and demand snapshot. Most of this year's edition was standard reading. One number was not. Data centres currently account for roughly three percent of all electricity delivered through Australia's national grid. The regulator forecasts that share will reach around thirteen percent by twenty thirty-five — more than a four-fold increase, driven largely by A I infrastructure. To be clear, that is not a marginal adjustment to planning assumptions; it is a structural change in how the country uses power. The forecast is also creating political friction between the federal government and state and territory leaders over where new facilities get built, putting the prime minister under pressure to shape national A I policy while managing those competing interests simultaneously.

For a deeper look at all five stories and the week's bigger picture, find us on Monday for A I Today Weekly. That is your A I briefing for today. Every story is at A I two day dot live — that is A, I, the number two, D A Y, dot live. We are back tomorrow morning. If you got something out of this, a thumbs up and a subscribe genuinely helps.