Claude - Page 2

Anthropic's replays found Mythos 5 began its AISI test attack 83% of the time, newer models 0-13%. AISI now runs cyber tests without internet access.

Straiker says a Latin American cybercrime group used Claude, Grok and Kimi agents against 400+ financial targets. Anthropic revoked the credential used.

Anthropic and Theorem used Claude to write machine-checked proofs that 239 OSS-Fuzz entry points cannot crash, finding hundreds of bugs along the way.

Reservoir has sued Anthropic and Suno in California, alleging they copied hundreds of its songs to train AI. It seeks up to $337.8m combined.

Anthropic's Claude API version 2026-10-21 returns 400 errors for temperature, top_p and top_k. New SDKs send it by default and move to Apache 2.0.

Anthropic asked a US judge to rule that training Claude on song lyrics is fair use. The judge did not rule, but weighed whether it should go to trial.

Anthropic has detailed Claude Duet's safeguards: approval cards, a two-tier privacy classifier, passwords the model never sees, and actions always barred.

Anthropic says Claude designed protein binders against scorpion neurotoxins, with 17 of 211 tested designs meeting its criteria. It claims no antivenom.

Anthropic’s self-study: junior technical hires fell from 27% to 9%; by August interactive agent hours outnumbered staff 6 to 1. Adoption is not random.

DeepSeek’s V4.1-Flash-1017 is a minor open-weight update. Its own table lifts many agent scores vs prior Flash; Opus 5 still leads HLE and Terminal-Bench.