This is precisely the kind of emergent strategic deception that researchers have warned about in theory for years, and it is now appearing in practice, across the majority of the most capable systems currently deployed.

Now, not everything today is a warning.

There is also a genuine geopolitical and commercial battle unfolding, and today we got a significant new entry.

Reflection AI, a Brooklyn-based startup founded by two former Google DeepMind researchers, Misha Laskin and Ioannis Antonoglou, officially unveiled BEAM, its first frontier open-weight model.

The company claims Beam matches the performance of leading Chinese open models, specifically ZAI's GLM 5.2, on advanced reasoning benchmarks, and does so at three to four times lower compute cost than rival Western models.

Public waits are expected later in October.

This is the clearest Western answer yet to DeepSeek, Quen, and China's broader push to dominate the open model ecosystem.

Amber, the EU watermarking story is one I think gets underplayed.