join.dev
AI news · 10 stories
Anthropic reports stronger defenses against Claude distillation · Anthropic details Scion, its in-house replacement for Git · Artificial Analysis adds a maintainer-focused coding leaderboard
Anthropic says new controls reduced reasoning data harvested from Claude by roughly 90–99% on its platform. The defenses restrict reasoning transfers and block extraction attempts, but require some integration changes and can affect legitimate requests.
Read at Anthropic
Anthropic described migrating its monorepo to Scion in five months as daily code merges rose from about 2,500 in May to over 30,000 in September. The engineering account shows how AI coding can move bottlenecks into source control and supporting infrastructure.
Read at Anthropic
Artificial Analysis launched independent results for Cognition’s FrontierCode 1.1 using 100 tasks and a common agent harness. Its leading model passed all blocking criteria in 58% of attempts, highlighting the gap between producing code and meeting repository maintainers’ requirements.
Read at Artificial Analysis
Intel disclosed D-series inference GPU specifications and said customer sampling is underway. Its air-cooled PCIe design targets existing data centers, with 160GB on Intel-branded cards and partner configurations supporting up to 480GB of memory.
Read at Intel
AMD introduced Helion, an open rack blueprint offering CPU-only and combined CPU/GPU configurations with a model-call router. Partner configurations are planned for November 2026, giving organizations more options for matching agent execution and inference to their infrastructure.
Read at AMD
NVIDIA outlined software updates ahead of RTX Spark laptop availability on October 16, including Nemotron 3.5 Bolt, PAIR and Broadcast 2.3. It schedules the Bolt model release for October 15, extending the ecosystem for running AI locally.
Read at NVIDIA
Around 400 scientists called for an international body to set containment standards and track significant incidents involving advanced cyber-capable AI. Their open letter advocates initial network isolation and independent oversight, arguing that voluntary commitments are insufficient.
Read at Reuters
Reuters reports that banking supervisors are increasing scrutiny of advanced AI, with US regulators questioning banks about model use. Sources point to cyber recovery, technology-provider dependence and rogue-agent risks as growing supervisory concerns.
Read at Reuters
The IMF’s October stability report warns that disappointing AI investment returns could trigger a market correction amplified by existing financial vulnerabilities. It identifies this as a potential risk alongside sovereign borrowing pressures and cyber incidents, rather than forecasting a crash.
Read at Anadolu Agency
IonQ agreed to buy Elisity in an all-stock deal worth up to $1.3 billion, with closing expected in the fourth quarter subject to approvals. Elisity’s network access controls cover AI agents as well as users and devices, extending IonQ’s enterprise-security offering.
Read at IonQ