Anthropic is embedding invisible watermarks in Claude text and images

Amazon Web Services (AWS) has integrated its AI-powered security platform, Continuum, directly into coding environments created by two major competitors: Anthropic's Claude Code and OpenAI's Codex. This strategic move was announced at Black Hat USA 2026 and represents AWS's most significant attempt to become the default security control plane for enterprise software development in the AI era.

The urgency behind this integration stems from a recent inflection point in enterprise security. Anthropic's Claude Mythos Preview, announced in April, demonstrated AI models with cybersecurity capabilities far surpassing prior systems. The model identified thousands of previously unknown zero-day vulnerabilities across major operating systems and web browsers, with a median time from discovery to weaponized exploit falling to under one hour by 2026.

This has significantly increased the vulnerability backlog, exacerbating the challenge faced by organizations.

AWS's Continuum platform is designed to address this vulnerability explosion by operating on an "agent-team loop architecture," automating the process of finding and fixing code flaws. The system comprises four phases: Discovery, where multiple frontier AI models scan code and ingest existing vulnerability backlogs; Prioritization, which contextualizes each finding against a customer's environment and business risk; Validation, where reproducible exploits are built in an isolated sandbox to confirm vulnerability exploitability; and Remediation, which offers tested fixes, whether configuration changes or code patches.

AWS is positioning Continuum as a single-bill security marketplace, absorbing the underlying token costs for the AI models used at each phase of the scan. The customer pays AWS a single price for Continuum, with AWS optimizing the AI model choice for each phase based on its effectiveness. This integration with Anthropic and OpenAI showcases AWS's commitment to becoming a comprehensive security control plane for enterprise software development, even in the face of competition.