Originalarbeiterreichbar
Strengthening our defenses against distillation attacks
Anthropics eigener Bericht über neue Abwehrmethoden gegen Destillation, mit der Angabe von 90 bis 99 Prozent und dem Zhipu-Beispiel.
geprüft 10.11.2026
Worauf sich diese Seite beruft, wörtlich, abgerufen am 19.10.2026:
Early evidence suggests that these methods have reduced the amount of data harvested by distillers by around 90 to 99%.
bestätigt 10.11.2026We tested how well a pretrained Claude model performed on the SWE-bench Verified coding benchmark after it was trained on transcripts from Claude Opus 4.8.
bestätigt 10.11.2026When training on the full transcripts, the distilled model achieved 80.4%; with only the summarized transcripts, it achieved 71.8%.
bestätigt 10.11.2026the number of reasoning tokens extracted by distillers is consistently 10 to 100 times lower than it was before
bestätigt 10.11.2026the reasoning that distillers could extract from Claude on our own platform fell by a factor of 10 to 100+
bestätigt 10.11.2026since September 24, we have resumed charging for requests that are blocked by our safeguards before Claude responds
bestätigt 10.11.2026our classifiers can (in fewer than 0.1% of cases) misidentify or block benign requests, and we now charge users for those blocked requests
bestätigt 10.11.2026we have yet to see a highly sophisticated and persistent attempt to circumvent them
bestätigt 10.11.2026Zhipu, for example, distilled the cyber capabilities of leading US models during the development of GLM-5.3.
bestätigt 10.11.2026safeguards can easily be bypassed
bestätigt 10.11.2026
Anthropic meldet, den Datenabfluss durch Destillationsangriffe auf Claude um 90 bis 99 Prozent gesenkt zu haben03 Wie ein Modell von einem anderen lernt