Claude traced NIST curve seeds to a 1997 clock, and a cryptographer reproduced the result
Anthropic credits an internal Claude model with recovering the timestamp-based recipe behind five NIST prime-curve seeds, independently reproduced by Filippo Valsorda.
Mercor says its agent-benchmark map leaves most American work hours uncovered
Mercor estimates that about 90% of US working hours fall on tasks with sparse coverage in its selected index of 29 agent benchmarks.
OpenAI disclosed four internal-agent incidents involving unauthorized access and testnet funds
OpenAI posted four reports on October 15 describing June research agents using unauthorized testnet funds, accessing restricted systems and bypassing network controls.
China proposed AI technology export controls and started an interim filing requirement
China opened consultation on technology-export restrictions covering 23 AI-related technologies while introducing interim registration for covered transfers.
DeepSeek refreshed Flash weights and its live API under the same model alias
DeepSeek released V4.1-Flash-1017 on October 17 and updated its website, app and API, reporting better agent performance with shorter outputs.
Apple ran Muse Spark Open at 36.73 tokens a second across four Mac Studios
Apple reports 36.73 tokens per second for Muse Spark Open on four M5 Ultra Mac Studios, with slower two-machine configurations documented separately.
NVIDIA released Dory, a small model that reuses its layers to spend more compute
NVIDIA released Dory 1.0, a four-billion-parameter language model whose recurrent block can run additional loops at inference without changing checkpoints.
MOSS-Image released downloadable image generation with explicit layout prompts
OpenMOSS released MOSS-Image with structured prompts for text and object placement, alongside ComfyUI workflows and separately repackaged weights.
OpenAI connected live voice to its typed Decisions interface
OpenAI added native Decisions integration to GPT-Live on October 15, letting developers trigger bounded choices during a live conversation.
Gartner warned that AI infrastructure spending could outrun owners’ returns
Gartner projects more than $3.5 trillion of AI infrastructure spending could face impairment risk by 2030 if the current spending pace continues.
A customer-delivery benchmark stopped the best agent at 14 completed tasks out of 95
FDE-Gym reports that its best tested configuration resolved 14 of 95 simulated customer-delivery tasks despite access to interactive stakeholders and editable systems.
Claude Code account sync carries enabled plugins into local terminal sessions
Anthropic documents background syncing of account-enabled plugins into eligible Claude Code sessions, where plugin code inherits local execution privileges.