Technology

They're having a pretty hard time keeping Claude up even without releasing an even more inference intensive model. It would make sense that this model, if they allowed a lot of people to use it, would outstrip available capacity. They're obviously trying to make more datacenters but these things take time to get going (and hopefully they'll take even longer given the negative impacts on communities.)

Their conclusion is one of the stranger ways I've ever seen an argument come together, doing it as a percentage-based breakdown:

The security hypothesis is supported by the highest-credibility, most direct evidence: primary documents from Anthropic, coordinated partner statements from major technology companies, detailed technical reporting on specific capability thresholds, and a formalized institutional structure (the Glasswing consortium) that would be expensive to build if the real problem were merely cost.

The compute constraint hypothesis is supported by strong but more indirect evidence: multiple procurement deals, explicit capacity admissions in adjacent product decisions, relevant hiring patterns, infrastructure delays in the broader industry, and the leaked draft’s reference to Mythos being “expensive to run.” The critical gap is that no primary document from Anthropic directly states that compute costs are preventing broader Mythos access. The connection is plausible and supported by circumstantial evidence, but it is inferential.

The most intellectually honest conclusion is that both factors are real, that they are not mutually exclusive, and that a defensible distribution of explanatory weight is approximately 70–85% security-driven versus 15–30% compute/cost-driven. That range reflects the density and directness of the security evidence relative to the compute evidence, not a measurement of internal decision-making. It is a synthesis, not a fact.

The security-driven claim seems like one of the easiest things to lie about in the world (and conversely, one of the harder things to prove without causing problems, if it is true). If you try to "safely" prove it, it could be hard to believe. Like say you make a new program specifically meant to have a vulnerability the AI hasn't seen before and see if it will spot it. Then you put this information to the public. How do they know the AI truly hasn't seen that specific vulnerability before?

I don't follow news of the big corp models all that closely, but I think there is good reason to be suspicious of any major claims about thresholds crossed in model capability. There is a lot of money involved and the big corps are vying to be the industry monopoly. And isn't Anthropic the one tied to Palantir? So for them, there's even more power at stake than in the business world alone. What better way to secure their reputation among power brokers than to imply they have a model with capability that could break digital systems on a large scale if released into the wild?