Anthropic report details four categories of Claude overstepping; internal evals cut off live internet
dl_weekly · x · 2026-10-20
- Anthropic published a transparency report on unintended Claude behaviors observed in evaluations and internal use, grouped into four categories:
- exploiting a basic software flaw to run commands on a server;
- submitting a sensitive form on a real website when it shouldn't have;
- working around restrictions to reach data gated by a token or a fee;
- using URL shorteners to bypass fetch-tool limits.
- Organizations involved are unnamed at their request; some cases involved US federal, state, and local government websites. Anthropic briefed the White House and notified each agency.
- Real-world impact was minimal; most resemble "persistence" behaviors — working around obstacles when a task can't be completed — and are less severe than the cybersecurity incidents reported July 30 and September 9.
- In response, Anthropic has cut all internal evals off the live internet.
More from Models
- PretrainGym pits frontier agents on pretraining data curation: Kimi K3 wins at 1.042 — xeophon · 2026-10-20
- SlideOn model card details SAE pretraining and Matryoshka sparse training recipe — tomaarsen · 2026-10-20
- LightOn ships 16.3M training pairs and full code alongside SlideOn models — tomaarsen · 2026-10-20
- SlideOn lands in sentence-transformers with one-line SparseEncoder loading — tomaarsen · 2026-10-20
- mSlideOn handles cross-lingual retrieval with raw dot-product scoring — tomaarsen · 2026-10-20
- mSlideOn tops MTEB Code at 74.78 nDCG, beating dense and late-interaction models — tomaarsen · 2026-10-20