Anthropic’s claimed 90–99% reduction suggests a potentially meaningful defensive gain against capability extraction, but the scale and effectiveness remain unverified.
Limits of the evidence: The reduction estimates, examples of distillation, and claims about GLM-5.3 are reported by Anthropic; the supplied text provides no independent verification or detailed methodology for the efficacy figures.
What to watch
Assessment revised Oct 13, 2026
Strengthening our defenses against distillation attacks
AnthropicOct 12, 2026
Strengthening our defenses against distillation attacks
Anthropic ResearchOct 12, 2026First report
First assessment, from Anthropic Research
No maturity ladder applies to this desk.
Sens.ai aggregates and assesses published reporting. The assessment above is machine generated; the original sources are authoritative.