AI SIGNAL ▮

Sunday, Oct 11, 2026

Top stories

Anthropic's updated usage policy bans sustained cruelty toward Claude, effective November 12

Anthropic published an updated usage policy that takes effect November 12, 2026. Its headline addition prohibits sustained and needless abusive or cruel behavior toward Anthropic's models; reporting says ordinary frustration, pushback, dark creative themes and model testing are not covered. The policy also spells out rules on deceptive influence campaigns and artificial activity, weapons development, and surveillance without consent, and requires that a qualified person be able to stop equipment when AI drives physical actions. Anthropic's page does not itemize what changed from the previous version, so the diff comes from press coverage.

Sources & depth

Ollama v0.40.3 pauses background model upgrades that slowed high-volume use

Ollama's v0.40.3 release candidate halts the background upgrade of old model files that arrived in v0.40.2, because it added overhead to every request in high-volume workloads such as embeddings. It also removes two spurious warnings from ollama launch claude (about Claude connectors) and ollama launch codex (about service tiers). It is tagged as a pre-release; anyone who saw latency regressions after v0.40.2 should watch for the final build.

Sources & depth

llama.cpp ships a run of builds: faster NVIDIA Vulkan matmul, vectorized fp16 conversion

Builds b11543 to b11552 tune Vulkan mixture-of-experts matrix multiplication to use four rows on NVIDIA GPUs (RDNA3/4 policy unchanged), vectorize fp32-to-fp16 conversion on CPU, fix multimodal input handling in speculative decoding, and add a LLAMA_ARG_SLOT_SAVE_PATH environment variable for the slot-save path. Routine maintenance; no benchmark numbers were published with the changes.

Sources & depth
Also today
Frontier releases
Hardware
From the community