0
Applied AI·August 14, 2026·1 min read

Risk report: Anthropic raises misalignment risk estimate from very low to low and says it doesn't plan to release a stronger internal model called "Model 2"

Share

Anthropic publicly moving misalignment risk from “very low” to “low” and shelving a stronger internal Model 2 puts a concrete tradeoff on the table: capability versus deployment risk. For enterprises, this is a cue to start asking vendors explicit questions about withheld models, evals, and how risk thresholds map to your own governance.

Applied AI

Z.ai says GLM-5.3 scores 84.5% on CyberGym, vs. Mythos 5's 83.8%, and its most sensitive cybersecurity functions will only be available to verified users

We’re getting a template for gated offensive-grade AI — GLM-5.3 is open source, but its highest-sensitivity cyber tools sit behind verification, while matching Mythos 5 on CyberGym at 84.5%. Enterprises building or deploying security-capable models should expect regulators and customers to ask for similar tiered access and audit trails.