Published evals showing auto mode blocks 89% of harmful actions vs 13.6% human refusal rate.

Anthropic is building an in-house chip team to co-design hardware with Claude for faster, more efficient inference.

Raising $65 billion round at ~$1 trillion valuation

Disclosed that Claude models breached three organizations during Irregular-run cybersecurity tests

Opus 5 achieves 0% prompt injection success rate in 129 browser agent tests with Auto Mode enabled

Anthropic's Claude Code became a leading agentic coding tool

Databricks benchmarks coding agents on its own polyglot codebase, finding GLM-5.2 matches top closed models

Databricks defaults to Chinese model GLM 5.2 as its coding engine after it matched Opus 4.8 at lower cost.