I love to see the doors to these capabilities opening up a bit more. Until recently this level of access sat with a small group running critical software, picked on terms that we outsiders couldn't see. Now the criteria are published, and smaller security firms, regional service providers and individual researchers can qualify. That puts better security tools in the hands of the good actors. In the earlier programme, Anthropic's partners found at least 129,000 verified vulnerabilities in four months. I'd want the people defending your systems using tools like this before attackers get something similar.

This is a sharp acceleration towards interfaces generated by AI on the spot, now in front of 1.2 billion weekly ChatGPT users, and they are likely to replace browsing traditional websites far sooner than most companies expect. For a company that wants to stay in control of how its brand shows up, it is crucial to be able to expose a single coherent view of products, pricing, content and assets to AI. That is harder than it sounds. Agents don't read the way people do. They filter typed data against constraints, so a product whose specs sit in a PDF, or that is named three different ways across retailers, is invisible to the model or has its data used while a rival with cleaner metadata gets the credit.

Palantir Foundry has already established itself across the globe as a powerful platform for pattern detection at both scale and pace. Tracking goods crossing the border is one of the hardest data problems a government has. It is a constant arms race where the adversaries identify and capitalise on the gaps in the system. Keeping up requires a data platform that allows you to rapidly integrate a myriad of data sources and reorientate your analysis to spot new behaviours, all while respecting governance and privacy controls.