Anthropic AI Updates: October 23, 2026

1. Full Claude Platform Now Generally Available on Google Cloud as an Anthropic-Managed Offering

Anthropic. Anthropic made Claude Models on Google Cloud (Anthropic managed) generally available, giving Google Cloud customers the native Claude API, including Managed Agents (beta), code execution, computer use, Agent Skills, the Files API, the MCP connector, prompt caching, and batch processing, with new features and betas shipping the same day as on the Claude Platform. Access uses existing Google Cloud IAM credentials, billing runs through a single Google Cloud invoice, and usage draws down existing Google Cloud commitments; customers also get the Claude Console. Fast mode for Claude Opus 5.5 (up to 2.5x faster output, $8 input and $40 output per million tokens) is also available, with US and Global endpoints at launch. Anthropic acts as data processor in this offering, while the existing Google-managed offering remains for customers who need Google as processor, strict regional data residency, or integrations such as Model Armor and Agent Engine. Source

2. Anthropic Donates the Agent Skills Standard to the Agentic AI Foundation

Anthropic. Anthropic is donating Agent Skills, the SKILL.md-based open standard for packaging agent knowledge and workflows, to the Agentic AI Foundation (AAIF) under the Linux Foundation, which already hosts the Model Context Protocol. The donation covers the Agent Skills name and logo, the GitHub organization, and the agentskills.io domain; Anthropic says more than 70 agent platforms, including ChatGPT, Cursor, GitHub Copilot, and VS Code, have integrated the standard, with 1 million skills published and nearly 280 million installed. Engineers from Anthropic, Google, OpenAI, and other companies will maintain it as individuals, with David Soria Parra and Mahesh Murag as Lead Maintainers, and the proposed roadmap covers versioned releases, a formal proposal process, and security guidance for skill authors and installers. Source

3. Claude Proves No-Crash Properties for 123 Open-Source Libraries at the Assembly Level

Anthropic. Anthropic’s Frontier Red Team, working with Theorem, reported that Claude proved that 239 fuzz-test entry points across 123 OSS-Fuzz libraries never crash on any input up to 64 MB, with proofs operating on compiled RISC-V binaries and checked by the Rocq proof kernel. The proofs total about 180 million lines of Rocq (a typical proof is about 600,000 lines for binaries with a median of 18,000 instructions) and assume a single-threaded static binary on RISC-V Linux. While proving, Claude found and patched hundreds of bugs, including an out-of-bounds write in a decompressor triggered only by rare 20 MB inputs and a rediscovery of CVE-2026-56132 in libexpat. Anthropic is publishing a subset of the proofs plus the RISC-V and OS semantics, and Theorem is releasing a Claude skill and offering to run the pipeline for other open-source projects. Source

4. Anthropic Opens a Call for Life Sciences Guest Investigators

Anthropic. Anthropic invited principal investigators at academic institutions, nonprofits, and government labs to propose 8 to 12 week projects using Claude on causal human biology, structural and computational molecular biology, medicinal chemistry, drug discovery, and clinical science. Guest investigators work directly with Anthropic researchers, who run the models and harnesses at no cost, and outputs, data, and code are to be published and open-sourced, with investigators retaining rights to their data and to independent follow-on research. Applications are open for about 14 days, with selected participants notified roughly 14 days after they close. Source

5. Claude Platform Guidance on Folding Subagents Into a Single Agent Loop

Anthropic. Anthropic’s Applied AI team published guidance and a migration kit for Claude Platform builders on when to replace routers, planners, and domain subagents with a single agent loop that progressively discloses context through skills and tools through tool search. Across several enterprise deployments, Anthropic reports that a single agent with skills frequently outperformed both a one-prompt design and a subagent design on quality, often at lower cost and latency, with one customer seeing roughly 2x faster runtime. The post still recommends subagents when context bloat measurably degrades performance, when parallel tasks need fresh context windows, or when one agent struggles with tool selection or conflicting instructions. Source