AI2Day Daily Brief — 18 Aug 2026

Good morning, it is Tuesday 18 August, and we are opening with a surgical robot doing open-heart bypass work through cuts between the ribs, an AI security model about to go public for free, and Nvidia committing over a hundred billion dollars to a single data center site.

Coronary artery bypass surgery is one of the most common cardiac operations in the world, and it almost always means sawing through the breastbone and weeks of painful recovery. Surgeons at El Camino Health Heart and Vascular Institute in Mountain View, California have started doing it a different way. Using Intuitive Surgical's da Vinci robotic system, Dr. Nels Carroll operated through small cuts between the ribs. His patient, sixty-four-year-old Raul Martin, was walking the hospital corridors the next day and went home within a couple of days. Robotic bypass accounts for fewer than one percent of these operations globally. The barriers are surgeon training, habit, and cost rather than outcomes. This case is one data point, but it illustrates what the gap between current practice and what the technology already allows looks like.

Stark gap. Next, a free-to-download AI model built for finding software vulnerabilities is two weeks away from public release. Z dot ai, a Chinese AI company, announced G L M five point three last Friday. It is an open-weight model, meaning you download it and run it on your own hardware rather than paying per query. According to Z dot ai's own testing, it matched or beat models from Anthropic and OpenAI on at least one major cybersecurity benchmark called CyberGym. The company also released a code-scanning service called OpenVuln built on top of it. The model rolls out first to trusted security partners, then broadly within two weeks. Open-weight cybersecurity tools lower the cost of both defence and attack, and who benefits more depends entirely on who moves faster.

On that note, a cautionary example of AI identification going wrong in a mundane setting. Matt Arnold, forty-six, walked into a Sainsbury's supermarket in the UK and was ejected by staff after the store's face-scanning system flagged him as a known shoplifter. He had not shoplifted. The system is called Facewatch. It captures an image of every person entering the store and compares it in real time against a private watchlist of people previously caught or suspected of theft. When it finds a match, it alerts staff. Arnold told The Guardian he felt embarrassed, mortified, and powerless. Sainsbury's attributed the incident to human error rather than the technology itself and has paused Facewatch at that location while it investigates. The episode is a practical illustration of what a false positive from an AI identification system costs a real person.

Shifting to enterprise AI adoption, AMD set itself a target of twenty-five percent productivity improvement from AI in software development over two to three years. It hit thirty percent in twelve months. Those figures come from IEEE Spectrum and use a specific definition: code that AI writes, clears all reviews, and ships in the final product. By that measure, more than one in five lines of AMD's production code now comes from an AI tool, and in some parts of the codebase the figure exceeds eighty percent. The clearest progress has been in graphics driver software, where AI agents went from automatically resolving six percent of reported bugs to more than seventy-five percent. AMD's stated next step is agent swarms, groups of AI systems working in parallel, with engineers setting goals rather than prescribing each step.

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.

And the biggest infrastructure story of the morning. Nvidia announced Monday it is investing one point five billion dollars in SB Energy, the data center developer backed by SoftBank and OpenAI. On top of that equity stake, Nvidia will extend up to one hundred and five billion dollars in credit to help build OpenAI's Ports-Pike data center near Cincinnati, Ohio. The facility is designed to start at four point two five gigawatts of power capacity and eventually reach eight gigawatts. To power it, a thirty-three billion dollar natural gas plant is planned on former U S Department of Energy land. BloombergNEF notes that natural gas plant construction costs have risen sixty-six percent over the last two years, and analysts have flagged that new demand at this scale could compete with export markets and push up gas prices more broadly. Nvidia also becomes the sole hardware supplier at the site, so the investment ties its revenue directly to the facility's build-out.

For the deeper look at all five of these stories and the week's other major developments, join us Monday for AI Today Weekly. That is your AI 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.