Claude Security Now Scans Continuously and Tracks Fixes
Claude Security watches connected repositories after a baseline scan and rescans changes based on risk. Findings are marked fixed when the fix ships, whether it came from Claude Code or elsewhere, and reopen if the bug returns. Each project gets an editable threat model, severity ratings, and attack-path explanations.
It is a public beta for Claude Enterprise with a new API, webhooks, an MCP connector, role-based access, and multi-repo projects. Eligible orgs get $25,000 in monthly credits for two months, through December 31. Anthropic says thousands of orgs have found over a million vulnerabilities.
Claude Designs Protein Binders Against Scorpion Neurotoxins
Per Anthropic’s Science Blog, via GIGAZINE, Claude worked autonomously for about 24 hours with protein design and folding tools, needing four follow-up instructions. It generated roughly 318,000 designs and proposed 180 candidates. With Claude’s help, the team added 31 more.
A partner lab tested 211 designs and 17 met the binding criteria. These were toxin-gene tests only, with no cell or animal work, so Anthropic does not claim an antivenom. All designs and binding data are on Hugging Face.
Anthropic-Managed Claude Launches on Google Cloud
Per the Claude Platform release notes, the Claude API now runs on Google Cloud as an Anthropic-managed service. Access is controlled by Google Cloud IAM and billing goes through Google Cloud Marketplace, so it fits into existing GCP spend.
Fast mode for Claude Opus 5.5 is available there too. Fast mode left research preview on the Claude API on October 15, and the header is no longer required.
New SDKs Switch to Apache 2.0, Profiles Gain Enforcement Fields
Alongside API version 2026-10-21, the new SDKs and ant CLI 2.0 move from the MIT license to Apache 2.0. Releases include Python 2.0.0, TypeScript 1.0.0, Go 2.0.0, Java 3.0.0, Ruby 2.0.0 and more, with breaking changes of their own.
User profiles now expose enforcement_status and enforcement_actions, showing whether Anthropic has restricted or banned an end customer. You can filter on open restrictions when listing profiles.
Replit Makes OpenAI Its Default, but Claude Stays Available
TechCrunch reports Replit CEO Amjad Masad said at Disrupt that Replit is no longer working with Anthropic, citing cost, partnership and performance. He later clarified that users can still pick Claude, but OpenAI is now the default for Replit Agent.
An anonymous insider disputes the framing and says the partnership is still in place. Since most users keep defaults, Anthropic revenue from Replit could still drop. Treat the exact scope as unsettled.
Analysts: Duet Is Restrained, Enterprises Are Not Ready
TechTarget quotes Forrester’s William McKeon-White saying nothing about Duet is especially unique except the restraint of its rollout, which he calls the better approach for agents this autonomous.
GAI Insights’ Paul Baier says companies over 1,000 employees do not look eager to deploy Duet, Dots or Muse soon, with IT, governance and security teams wary of confidential data flowing through them.
Agents Are Moving Into the Stack
Look at the week. Security scanning that never sleeps. Claude running inside your cloud account. An agent doing 24 hours of lab design. Duet in your inbox. The pattern is autonomy with a paper trail: threat models, IAM, approval gates, open datasets.
The analyst quotes explain why. Capability is not the bottleneck anymore. Getting security and IT to say yes is. Replit’s wobble is the other lesson: defaults win, and rivals are fighting for them.