In April 2026, Anthropic built its most capable model to date and then chose not to ship it. Mythos went to a small set of named partners; everyone else was locked out. That single decision is what the AmChamSG AI Advantage panel came to examine, and our CEO Roger Olofsson joined speakers from CrowdStrike, Microsoft, Cisco, and Allen & Gledhill to work through it. Our view going in, and coming out, is that a curated frontier changes who can defend, who can build, and above all who you need to hire.
Why a model that was never released mattered so much
This was the first time a frontier lab openly said it had built something and would not release it. Mythos sits with around fifty named partners while the rest of the market waits. The panel's starting question was what that does to a company that is not on the list. Capability stops being something you buy and starts being something you are granted, and that shifts the competitive question from budget to relationships, trust, and the credibility of your own security and governance posture.
The security asymmetry
The uncomfortable part of Mythos is that the same capability that finds vulnerabilities at scale can exploit them at scale. Defensive programmes built on quarterly cycles and manual review do not survive contact with that. What the conversation kept returning to is that this is not solved by procurement. It is solved by people: security leaders who understand model behaviour, engineers who can operate agentic tooling safely, and executives who can make risk calls quickly and defensibly.
Back to the mainframe, at speed
In the 1960s only a handful of institutions could reach frontier compute. The curated frontier points in a similar direction, compressed into years rather than decades, with real physical constraints on chips, data centres, and energy behind it. We do not read that as a reason for smaller companies to give up. We read it as a reason to be sharper about leverage: a small, elite, AI-augmented team that has genuine access and genuine judgment can outperform a large one that has neither.
What this means for hiring
Our contribution to the panel was the talent lens the others were pointing at without naming. Every scenario on that stage — partner access, autonomous defence, regulated deployment, compute economics — resolves into a hiring problem. Companies need leaders who can hold technical depth and governance credibility at once, and there are not many of them. This is exactly the search we run through the Olofsson AI Lab and our AI-powered talent engine, which lets us map, assess, and reach that thin layer of the market far faster than a keyword search ever could.