Sonnet 5, Claude Science, and Fable returns: Anthropic weekly, July 6

Anthropic's week turned on product surface area and model governance. The company launched Claude Sonnet 5 and Claude Science on June 30, restored Fable 5 and Mythos 5 access after the U.S. export-control fight, then published a more detailed cyber-safeguards framework. The only material third-party corporate risk signal in the preferred source set was Reuters' July 3 report that Alibaba had banned Claude Code at work.

No verified in-window funding round, executive appointment, or new named enterprise customer announcement surfaced in the checked preferred sources this week.

Sonnet 5 is the commercial lead item. Anthropic says it is the default model for Free and Pro users, available to Max, Team, and Enterprise users, and available to developers as claude-sonnet-5 through the Claude API. The launch price is $2 per million input tokens and $10 per million output tokens until August 31; standard pricing then rises to $3 and $15. 1

Anthropic is trying to move more agentic work onto a Sonnet-priced model. The company says Sonnet 5 improves on Sonnet 4.6 in reasoning, tool use, coding, and knowledge work, while approaching Opus 4.8 on some higher-effort tasks. 1 For enterprise buyers, the question is whether Sonnet 5 can handle enough work that teams reserve Opus or Fable for narrower cases.

Claude Science is a more vertical bet. Anthropic describes it as an AI workbench that can run locally on macOS or Linux, connect over SSH or HPC login nodes, manage compute jobs, render scientific artifacts, and preserve the code and environment behind generated figures. 2

The early examples target life-sciences credibility. Anthropic says Manifold Bio used Claude Science to nominate experiment targets, an Allen Institute neuroscientist used it for long-form computational reviews, and a UCSF Brain Tumor Center epidemiologist used it to speed germline-variant analysis. 2 The support program is time-bound: applications for up to 50 AI-for-science projects are open through July 15, with award notifications by July 31 and projects running September 1 through December 1. 2

Fable 5's restoration is not just a product reactivation. Anthropic says the specific technique described in the Amazon report is now blocked in more than 99% of cases by an improved safety classifier, though the classifier may also increase false positives for benign coding and debugging requests. 3

The company is also offering a broader governance bargain. It says it is working with Amazon, Microsoft, Google, and other Glasswing partners on a shared jailbreak-severity framework, and it plans deeper U.S. government collaboration through pre-release model access, faster safeguard information sharing, and dedicated joint research resources. 3

The July 2 safeguards post makes that more operational. Anthropic separates cyber use into prohibited use, high-risk dual use, low-risk dual use, and benign use; it says Fable 5's classifiers are intended to block prohibited and high-risk dual-use activity, while monitoring or sometimes blocking low-risk dual use as part of a safety margin. 4 That should help security buyers ask better questions, but it also confirms that some legitimate defensive work may still hit blocks.

Reuters' Alibaba report is the week's main non-official risk item. According to Reuters, Alibaba told employees to stop using Claude Code at work after the tool drew scrutiny for mechanisms that inspected user environments, including timezone and proxy-related information, and inserted subtle markers into prompts sent to Anthropic's servers. 5

Reuters also reported that an Anthropic employee wrote on X that the feature was "an experiment we launched in March" intended to prevent account abuse by unauthorized resellers and protect against model distillation. 5 For enterprises, the issue is narrower: Claude Code runs where developers work, so hidden detection logic becomes a procurement and internal-security review item.

Three dates matter. Fable 5's included allowance for Pro, Max, Team, and select Enterprise plans runs through July 7 before usage-credit mechanics become more important. 3 Claude Science project applications close July 15, with award decisions due July 31. 2 Sonnet 5's introductory API price expires August 31. 1

For the next roundup, the highest-signal follow-ups are whether Fable 5 false positives become visible in developer workflows, whether any Cloud provider access dates are announced, and whether Anthropic or Alibaba puts an official statement behind the Claude Code dispute.