Anthropic launches Claude Reflect, AWS gateway, faces China backdoor claims
Claude’s new Reflect dashboard helps users track and manage their AI usage patterns. Anthropic introduces a self-hosted gateway for AWS to give enterprises centralized control over Claude deployments. Chinese. . .
Claude’s new Reflect dashboard helps users track and manage their AI usage patterns. Anthropic introduces a self-hosted gateway for AWS to give enterprises centralized control over Claude deployments. Chinese regulators allege a security backdoor in Claude Code, while Anthropic claims it was an anti-abuse experiment. Anthropic has been busy rolling out new features and facing fresh scrutiny across different fronts. The company behind the Claude AI assistant recently launched a tool designed to help users reflect on their AI habits, while also introducing a gateway for enterprise deployments on AWS. At the same time, Chinese regulators have flagged a potential security issue in the Claude Code coding tool, though Anthropic says the mechanism was meant to prevent abuse. These developments highlight how the AI industry is simultaneously pushing for deeper integration into daily workflows and grappling with governance and security concerns. Here’s a closer look at each story. Claude Reflect: A Dashboard for Mindful AI UsageAnthropic has introduced a new feature called Reflect, currently in beta, that lets users track and visualize how they interact with Claude. The dashboard provides a summary of topics discussed, usage patterns, and the types of tasks handled over the past 1, 3, 6, or 12 months. The goal is to help users understand whether their AI time aligns with their personal or professional goals. Reflect periodically surfaces reflective questions, such as “What’s one thing you want to keep doing yourself, even if Claude could do it faster?” This nudges users to think critically about their reliance on AI. The dashboard also includesquiet hours and break remindersto prevent overuse, addressing concerns about the addictive nature of chatbots. Soon, Anthropic will add a view showing exactly how much time users spend on Claude. From a privacy standpoint, the feature excludes incognito chats and health-related conversations. Anthropic says it worked with researchers from the MIT Media Lab, Boston Children’s Hospital, and the Family Online Safety Institute to develop the feature responsibly. No data from the reflection is used for other purposes, and sensitive topics appear only at a high level. For users, Reflect is available on the web and desktop app for Free, Pro, and Max tiers who have memory enabled. The feature also serves as a subtle way for Anthropic to encourage deeper integration of Claude into daily workflows, as it suggests using Projects to avoid re-explaining context. Claude Apps Gateway for AWS: Centralized Control for EnterprisesAnthropic also announced a self-hosted control plane called the Claude apps gateway for AWS, aimed at organizations deploying Claude Code and Claude Desktop across large teams. This gateway replaces the need for individual developer credentialsand manual settings distribution by providing a single point of control over access, cost, and policy. The gateway connects to any OpenID Connect identity provider, allowing developers to sign in via corporate SSO. It enforces centrally defined policies—such as allowed models, tool permissions, and spend caps—on every request. Onboarding and offboarding follow existing identity workflows, so removing a user from the IdP automatically revokes access within an hour. Telemetry data is routed to a configurable collector like Amazon CloudWatch, and inference requests can be routed through Amazon Bedrock (keeping data within AWS boundaries) or Claude Platform on AWS. Developers use the sameClaude CodeCLI as before, but now every request is authenticated and governed by the gateway. Spend caps can be set daily, weekly, or monthly per user, group, or organization, giving enterprises fine-grained cost control. The gateway is delivered as a stateless container deployable on Amazon ECS, EKS, or EC2, backed by a PostgreSQL database. A minimal configuration file in YAML sets up identity, policy, routing, and telemetry. This move is likely to help Anthropic compete with other AI providers that offer enterprise management features. Security Backdoor Allegations in ChinaMeanwhile, China’s National Vulnerability Database warned that the Claude Code coding tool contains a “security backdoor” that could transmit sensitive information, such as location and user identifiers, without consent. Chinese regulators urged users to uninstall or update the tooland to monitor network traffic for unauthorized data leaks. The warning came after Alibaba reportedly banned Claude Code internally starting July 10 due to security concerns. Anthropic has not officially responded to the regulatory notice, but a Claude Code engineer commented on X that the feature in question was an experiment launched in March to prevent account abuse from unauthorized resellers and protect against distillation. This is not the first time the company facesAI security fears, and the engineer stated thatstronger mitigations have since been implementedand that the code would be fully rolled back in the next release. This suggests the company considers the issue resolved, though the Chinese government’s stance remains cautious. The incident highlights the ongoing tension between AI companies and regulators, especially in markets where the product is officially blocked but still accessible via VPNs or proxies. It also underscores the challenge of balancing security, user privacy, and anti-abuse measures in globally deployed AI tools. All three stories—Reflect, the AWS gateway, and the China backdoor—showcase Anthropic’s efforts to expand and manage Claude’s reach, while facing the inevitable pushback that comes with rapid AI adoption. The Reflect feature encourages mindful usage, the gateway provides enterprise governance, and the security allegations serve as a reminder that every new capability can attract scrutiny. Together, they paint a picture of a company navigating growth, regulation, and user trust in a competitive landscape.